diff options
Diffstat (limited to 'db2/hash')
| -rw-r--r-- | db2/hash/hash.c | 1151 | ||||
| -rw-r--r-- | db2/hash/hash_auto.c | 161 | ||||
| -rw-r--r-- | db2/hash/hash_debug.c | 92 | ||||
| -rw-r--r-- | db2/hash/hash_dup.c | 295 | ||||
| -rw-r--r-- | db2/hash/hash_page.c | 1088 | ||||
| -rw-r--r-- | db2/hash/hash_rec.c | 281 | ||||
| -rw-r--r-- | db2/hash/hash_stat.c | 37 |
7 files changed, 1524 insertions, 1581 deletions
diff --git a/db2/hash/hash.c b/db2/hash/hash.c index 0265f19659..0d202fce20 100644 --- a/db2/hash/hash.c +++ b/db2/hash/hash.c @@ -47,7 +47,7 @@ #include "config.h" #ifndef lint -static const char sccsid[] = "@(#)hash.c 10.45 (Sleepycat) 5/11/98"; +static const char sccsid[] = "@(#)hash.c 10.63 (Sleepycat) 12/11/98"; #endif /* not lint */ #ifndef NO_SYSTEM_INCLUDES @@ -64,23 +64,23 @@ static const char sccsid[] = "@(#)hash.c 10.45 (Sleepycat) 5/11/98"; #include "db_am.h" #include "db_ext.h" #include "hash.h" +#include "btree.h" #include "log.h" +#include "db_shash.h" +#include "lock.h" +#include "lock_ext.h" static int __ham_c_close __P((DBC *)); static int __ham_c_del __P((DBC *, u_int32_t)); +static int __ham_c_destroy __P((DBC *)); static int __ham_c_get __P((DBC *, DBT *, DBT *, u_int32_t)); static int __ham_c_put __P((DBC *, DBT *, DBT *, u_int32_t)); -static int __ham_c_init __P((DB *, DB_TXN *, DBC **)); -static int __ham_cursor __P((DB *, DB_TXN *, DBC **)); static int __ham_delete __P((DB *, DB_TXN *, DBT *, u_int32_t)); -static int __ham_dup_return __P((HTAB *, HASH_CURSOR *, DBT *, u_int32_t)); -static int __ham_get __P((DB *, DB_TXN *, DBT *, DBT *, u_int32_t)); -static void __ham_init_htab __P((HTAB *, u_int32_t, u_int32_t)); -static int __ham_lookup __P((HTAB *, - HASH_CURSOR *, const DBT *, u_int32_t, db_lockmode_t)); -static int __ham_overwrite __P((HTAB *, HASH_CURSOR *, DBT *)); -static int __ham_put __P((DB *, DB_TXN *, DBT *, DBT *, u_int32_t)); -static int __ham_sync __P((DB *, u_int32_t)); +static int __ham_dup_return __P((DBC *, DBT *, u_int32_t)); +static int __ham_expand_table __P((DBC *)); +static void __ham_init_htab __P((DBC *, u_int32_t, u_int32_t)); +static int __ham_lookup __P((DBC *, const DBT *, u_int32_t, db_lockmode_t)); +static int __ham_overwrite __P((DBC *, DBT *)); /************************** INTERFACE ROUTINES ***************************/ /* OPEN/CLOSE */ @@ -96,65 +96,53 @@ __ham_open(dbp, dbinfo) DB_INFO *dbinfo; { DB_ENV *dbenv; - DBC *curs; - HTAB *hashp; + DBC *dbc; + HASH_CURSOR *hcp; int file_existed, ret; + dbc = NULL; dbenv = dbp->dbenv; - if ((hashp = (HTAB *)__db_calloc(1, sizeof(HTAB))) == NULL) - return (ENOMEM); - hashp->dbp = dbp; - /* Set the hash function if specified by the user. */ if (dbinfo != NULL && dbinfo->h_hash != NULL) - hashp->hash = dbinfo->h_hash; + dbp->h_hash = dbinfo->h_hash; /* - * Initialize the remaining fields of the dbp. The type, close and - * fd functions are all set in db_open. + * Initialize the remaining fields of the dbp. The only function + * that differs from the default set is __ham_stat(). */ - dbp->internal = hashp; - dbp->cursor = __ham_cursor; + dbp->internal = NULL; + dbp->am_close = __ham_close; dbp->del = __ham_delete; - dbp->get = __ham_get; - dbp->put = __ham_put; - dbp->sync = __ham_sync; - - /* If locking is turned on, lock the meta data page. */ - if (F_ISSET(dbp, DB_AM_LOCKING)) { - dbp->lock.pgno = BUCKET_INVALID; - if ((ret = lock_get(dbenv->lk_info, dbp->locker, - 0, &dbp->lock_dbt, DB_LOCK_READ, &hashp->hlock)) != 0) { - if (ret < 0) - ret = EAGAIN; - goto out; - } - } + dbp->stat = __ham_stat; + + /* Get a cursor we can use for the rest of this function. */ + if ((ret = dbp->cursor(dbp, NULL, &dbc, 0)) != 0) + goto out; + + hcp = (HASH_CURSOR *)dbc->internal; + GET_META(dbp, hcp, ret); + if (ret != 0) + goto out; /* - * Now, we can try to read the meta-data page and figure out - * if we set up locking and get the meta-data page properly. * If this is a new file, initialize it, and put it back dirty. */ - if ((ret = __ham_get_page(hashp->dbp, 0, (PAGE **)&hashp->hdr)) != 0) - goto out; - /* Initialize the hashp structure */ - if (hashp->hdr->magic == DB_HASHMAGIC) { + /* Initialize the hdr structure */ + if (hcp->hdr->magic == DB_HASHMAGIC) { file_existed = 1; /* File exists, verify the data in the header. */ - if (hashp->hash == NULL) - hashp->hash = - hashp->hdr->version < 5 ? __ham_func4 : __ham_func5; - if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != - hashp->hdr->h_charkey) { - __db_err(hashp->dbp->dbenv, - "hash: incompatible hash function"); + if (dbp->h_hash == NULL) + dbp->h_hash = + hcp->hdr->version < 5 ? __ham_func4 : __ham_func5; + if (dbp->h_hash(CHARKEY, sizeof(CHARKEY)) != + hcp->hdr->h_charkey) { + __db_err(dbp->dbenv, "hash: incompatible hash function"); ret = EINVAL; goto out; } - if (F_ISSET(hashp->hdr, DB_HASH_DUP)) + if (F_ISSET(hcp->hdr, DB_HASH_DUP)) F_SET(dbp, DB_AM_DUP); } else { /* @@ -163,59 +151,27 @@ __ham_open(dbp, dbinfo) */ file_existed = 0; if (F_ISSET(dbp, DB_AM_LOCKING) && - ((ret = lock_put(dbenv->lk_info, hashp->hlock)) != 0 || - (ret = lock_get(dbenv->lk_info, dbp->locker, 0, - &dbp->lock_dbt, DB_LOCK_WRITE, &hashp->hlock)) != 0)) { + ((ret = lock_put(dbenv->lk_info, hcp->hlock)) != 0 || + (ret = lock_get(dbenv->lk_info, dbc->locker, 0, + &dbc->lock_dbt, DB_LOCK_WRITE, &hcp->hlock)) != 0)) { if (ret < 0) ret = EAGAIN; goto out; } - __ham_init_htab(hashp, - dbinfo != NULL ? dbinfo->h_nelem : 0, + __ham_init_htab(dbc, dbinfo != NULL ? dbinfo->h_nelem : 0, dbinfo != NULL ? dbinfo->h_ffactor : 0); if (F_ISSET(dbp, DB_AM_DUP)) - F_SET(hashp->hdr, DB_HASH_DUP); - if ((ret = __ham_dirty_page(hashp, (PAGE *)hashp->hdr)) != 0) + F_SET(hcp->hdr, DB_HASH_DUP); + if ((ret = __ham_dirty_page(dbp, (PAGE *)hcp->hdr)) != 0) goto out; } - /* Initialize the default cursor. */ - __ham_c_init(dbp, NULL, &curs); - TAILQ_INSERT_TAIL(&dbp->curs_queue, curs, links); - - /* Allocate memory for our split buffer. */ - if ((hashp->split_buf = (PAGE *)__db_malloc(dbp->pgsize)) == NULL) { - ret = ENOMEM; - goto out; - } - -#ifdef NO_STATISTICS_FOR_DB_ERR - __db_err(dbp->dbenv, - "%s%lx\n%s%ld\n%s%ld\n%s%ld\n%s%ld\n%s0x%lx\n%s0x%lx\n%s%ld\n%s%ld\n%s0x%lx", - "TABLE POINTER ", (long)hashp, - "BUCKET SIZE ", (long)hashp->hdr->pagesize, - "FILL FACTOR ", (long)hashp->hdr->ffactor, - "MAX BUCKET ", (long)hashp->hdr->max_bucket, - "OVFL POINT ", (long)hashp->hdr->ovfl_point, - "LAST FREED ", (long)hashp->hdr->last_freed, - "HIGH MASK ", (long)hashp->hdr->high_mask, - "LOW MASK ", (long)hashp->hdr->low_mask, - "NELEM ", (long)hashp->hdr->nelem, - "FLAGS ", (long)hashp->hdr->flags); -#endif - /* Release the meta data page */ - (void)__ham_put_page(hashp->dbp, (PAGE *)hashp->hdr, 0); - if (F_ISSET(dbp, DB_AM_LOCKING) && - (ret = lock_put(dbenv->lk_info, hashp->hlock)) != 0) { - if (ret < 0) - ret = EAGAIN; + RELEASE_META(dbp, hcp); + if ((ret = dbc->c_close(dbc)) != 0) goto out; - } - hashp->hlock = 0; - hashp->hdr = NULL; /* Sync the file so that we know that the meta data goes to disk. */ if (!file_existed && (ret = dbp->sync(dbp, 0)) != 0) goto out; @@ -232,27 +188,8 @@ int __ham_close(dbp) DB *dbp; { - HTAB *hashp; - int ret, t_ret; - - DEBUG_LWRITE(dbp, NULL, "ham_close", NULL, NULL, 0); - hashp = (HTAB *)dbp->internal; - ret = 0; - - /* Free the split page. */ - if (hashp->split_buf) - FREE(hashp->split_buf, dbp->pgsize); - - if (hashp->hdr && (t_ret = __ham_put_page(hashp->dbp, - (PAGE *)hashp->hdr, 0)) != 0 && ret == 0) - ret = t_ret; - if (hashp->hlock && (t_ret = lock_put(hashp->dbp->dbenv->lk_info, - hashp->hlock)) != 0 && ret == 0) - ret = t_ret; - - FREE(hashp, sizeof(HTAB)); - dbp->internal = NULL; - return (ret); + COMPQUIET(dbp, NULL); + return (0); } /************************** LOCAL CREATION ROUTINES **********************/ @@ -260,408 +197,204 @@ __ham_close(dbp) * Returns 0 on No Error */ static void -__ham_init_htab(hashp, nelem, ffactor) - HTAB *hashp; +__ham_init_htab(dbc, nelem, ffactor) + DBC *dbc; u_int32_t nelem, ffactor; { + DB *dbp; + HASH_CURSOR *hcp; int32_t l2, nbuckets; - memset(hashp->hdr, 0, sizeof(HASHHDR)); - hashp->hdr->ffactor = ffactor; - hashp->hdr->pagesize = hashp->dbp->pgsize; - ZERO_LSN(hashp->hdr->lsn); - hashp->hdr->magic = DB_HASHMAGIC; - hashp->hdr->version = DB_HASHVERSION; - if (hashp->hash == NULL) - hashp->hash = - hashp->hdr->version < 5 ? __ham_func4 : __ham_func5; - hashp->hdr->h_charkey = hashp->hash(CHARKEY, sizeof(CHARKEY)); - if (nelem != 0 && hashp->hdr->ffactor != 0) { - nelem = (nelem - 1) / hashp->hdr->ffactor + 1; + hcp = (HASH_CURSOR *)dbc->internal; + dbp = dbc->dbp; + memset(hcp->hdr, 0, sizeof(HASHHDR)); + hcp->hdr->ffactor = ffactor; + hcp->hdr->pagesize = dbp->pgsize; + ZERO_LSN(hcp->hdr->lsn); + hcp->hdr->magic = DB_HASHMAGIC; + hcp->hdr->version = DB_HASHVERSION; + + if (dbp->h_hash == NULL) + dbp->h_hash = hcp->hdr->version < 5 ? __ham_func4 : __ham_func5; + hcp->hdr->h_charkey = dbp->h_hash(CHARKEY, sizeof(CHARKEY)); + if (nelem != 0 && hcp->hdr->ffactor != 0) { + nelem = (nelem - 1) / hcp->hdr->ffactor + 1; l2 = __db_log2(nelem > 2 ? nelem : 2); } else l2 = 2; nbuckets = 1 << l2; - hashp->hdr->ovfl_point = l2; - hashp->hdr->last_freed = PGNO_INVALID; + hcp->hdr->ovfl_point = l2; + hcp->hdr->last_freed = PGNO_INVALID; - hashp->hdr->max_bucket = hashp->hdr->high_mask = nbuckets - 1; - hashp->hdr->low_mask = (nbuckets >> 1) - 1; - memcpy(hashp->hdr->uid, hashp->dbp->lock.fileid, DB_FILE_ID_LEN); + hcp->hdr->max_bucket = hcp->hdr->high_mask = nbuckets - 1; + hcp->hdr->low_mask = (nbuckets >> 1) - 1; + memcpy(hcp->hdr->uid, dbp->fileid, DB_FILE_ID_LEN); } -/********************** DESTROY/CLOSE ROUTINES ************************/ - - -/* - * Write modified pages to disk - * - * Returns: - * 0 == OK - * -1 ERROR - */ static int -__ham_sync(dbp, flags) - DB *dbp; - u_int32_t flags; -{ - int ret; - - DEBUG_LWRITE(dbp, NULL, "ham_sync", NULL, NULL, flags); - if ((ret = __db_syncchk(dbp, flags)) != 0) - return (ret); - if (F_ISSET(dbp, DB_AM_RDONLY)) - return (0); - - if ((ret = memp_fsync(dbp->mpf)) == DB_INCOMPLETE) - ret = 0; - - return (ret); -} - -/*******************************SEARCH ROUTINES *****************************/ -/* - * All the access routines return - * - * Returns: - * 0 on SUCCESS - * 1 to indicate an external ERROR (i.e. key not found, etc) - * -1 to indicate an internal ERROR (i.e. out of memory, etc) - */ - -static int -__ham_get(dbp, txn, key, data, flags) +__ham_delete(dbp, txn, key, flags) DB *dbp; DB_TXN *txn; DBT *key; - DBT *data; u_int32_t flags; { - DB *ldbp; - HTAB *hashp; + DBC *dbc; HASH_CURSOR *hcp; - int ret, t_ret; + int ret, tret; - DEBUG_LREAD(dbp, txn, "ham_get", key, NULL, flags); - if ((ret = __db_getchk(dbp, key, data, flags)) != 0) - return (ret); + DB_PANIC_CHECK(dbp); - ldbp = dbp; - if (F_ISSET(dbp, DB_AM_THREAD) && - (ret = __db_gethandle(dbp, __ham_hdup, &ldbp)) != 0) + if ((ret = + __db_delchk(dbp, key, flags, F_ISSET(dbp, DB_AM_RDONLY))) != 0) return (ret); - hashp = (HTAB *)ldbp->internal; - SET_LOCKER(ldbp, txn); - GET_META(ldbp, hashp); - - hashp->hash_accesses++; - hcp = (HASH_CURSOR *)TAILQ_FIRST(&ldbp->curs_queue)->internal; - if ((ret = __ham_lookup(hashp, hcp, key, 0, DB_LOCK_READ)) == 0) { - if (F_ISSET(hcp, H_OK)) - ret = __ham_dup_return(hashp, hcp, data, DB_FIRST); - else /* Key was not found */ - ret = DB_NOTFOUND; - } - - if ((t_ret = __ham_item_done(hashp, hcp, 0)) != 0 && ret == 0) - ret = t_ret; - RELEASE_META(ldbp, hashp); - if (F_ISSET(dbp, DB_AM_THREAD)) - __db_puthandle(ldbp); - return (ret); -} - -static int -__ham_put(dbp, txn, key, data, flags) - DB *dbp; - DB_TXN *txn; - DBT *key; - DBT *data; - u_int32_t flags; -{ - DB *ldbp; - DBT tmp_val, *myval; - HASH_CURSOR *hcp; - HTAB *hashp; - u_int32_t nbytes; - int ret, t_ret; - - DEBUG_LWRITE(dbp, txn, "ham_put", key, data, flags); - if ((ret = __db_putchk(dbp, key, data, - flags, F_ISSET(dbp, DB_AM_RDONLY), F_ISSET(dbp, DB_AM_DUP))) != 0) + if ((ret = dbp->cursor(dbp, txn, &dbc, DB_WRITELOCK)) != 0) return (ret); - ldbp = dbp; - if (F_ISSET(dbp, DB_AM_THREAD) && - (ret = __db_gethandle(dbp, __ham_hdup, &ldbp)) != 0) - return (ret); + DEBUG_LWRITE(dbc, txn, "ham_delete", key, NULL, flags); - hashp = (HTAB *)ldbp->internal; - SET_LOCKER(ldbp, txn); - GET_META(ldbp, hashp); - hcp = TAILQ_FIRST(&ldbp->curs_queue)->internal; - - nbytes = (ISBIG(hashp, key->size) ? HOFFPAGE_PSIZE : - HKEYDATA_PSIZE(key->size)) + - (ISBIG(hashp, data->size) ? HOFFPAGE_PSIZE : - HKEYDATA_PSIZE(data->size)); - - hashp->hash_accesses++; - ret = __ham_lookup(hashp, hcp, key, nbytes, DB_LOCK_WRITE); - - if (ret == DB_NOTFOUND) { - ret = 0; - if (hcp->seek_found_page != PGNO_INVALID && - hcp->seek_found_page != hcp->pgno) { - if ((ret = __ham_item_done(hashp, hcp, 0)) != 0) - goto out; - hcp->pgno = hcp->seek_found_page; - hcp->bndx = NDX_INVALID; - } + hcp = (HASH_CURSOR *)dbc->internal; + GET_META(dbp, hcp, ret); + if (ret != 0) + goto out; - if (F_ISSET(data, DB_DBT_PARTIAL) && data->doff != 0) { - /* - * Doing a partial put, but the key does not exist - * and we are not beginning the write at 0. We - * must create a data item padded up to doff and - * then write the new bytes represented by val. - */ - ret = __ham_init_dbt(&tmp_val, data->size + data->doff, - &hcp->big_data, &hcp->big_datalen); - if (ret == 0) { - memset(tmp_val.data, 0, data->doff); - memcpy((u_int8_t *)tmp_val.data + data->doff, - data->data, data->size); - myval = &tmp_val; - } - } else - myval = (DBT *)data; - - if (ret == 0) - ret = __ham_add_el(hashp, hcp, key, myval, H_KEYDATA); - } else if (ret == 0 && F_ISSET(hcp, H_OK)) { - if (flags == DB_NOOVERWRITE) - ret = DB_KEYEXIST; - else if (F_ISSET(ldbp, DB_AM_DUP)) - ret = __ham_add_dup(hashp, hcp, data, DB_KEYLAST); + hcp->stats.hash_deleted++; + if ((ret = __ham_lookup(dbc, key, 0, DB_LOCK_WRITE)) == 0) { + if (F_ISSET(hcp, H_OK)) + ret = __ham_del_pair(dbc, 1); else - ret = __ham_overwrite(hashp, hcp, data); - } - - /* Free up all the cursor pages. */ - if ((t_ret = __ham_item_done(hashp, hcp, ret == 0)) != 0 && ret == 0) - ret = t_ret; - /* Now check if we have to grow. */ -out: if (ret == 0 && F_ISSET(hcp, H_EXPAND)) { - ret = __ham_expand_table(hashp); - F_CLR(hcp, H_EXPAND); + ret = DB_NOTFOUND; } - if ((t_ret = __ham_item_done(hashp, hcp, ret == 0)) != 0 && ret == 0) - ret = t_ret; - RELEASE_META(ldbp, hashp); - if (F_ISSET(dbp, DB_AM_THREAD)) - __db_puthandle(ldbp); + RELEASE_META(dbp, hcp); +out: if ((tret = dbc->c_close(dbc)) != 0 && ret == 0) + ret = tret; return (ret); } -static int -__ham_cursor(dbp, txnid, dbcp) - DB *dbp; - DB_TXN *txnid; - DBC **dbcp; -{ +/* ****************** CURSORS ********************************** */ +/* + * __ham_c_init -- + * Initialize the hash-specific portion of a cursor. + * + * PUBLIC: int __ham_c_init __P((DBC *)); + */ +int +__ham_c_init(dbc) + DBC *dbc; + { + HASH_CURSOR *new_curs; int ret; - DEBUG_LWRITE(dbp, txnid, "ham_cursor", NULL, NULL, 0); - if ((ret = __ham_c_init(dbp, txnid, dbcp)) != 0) + if ((ret = __os_calloc(1, sizeof(struct cursor_t), &new_curs)) != 0) + return (ret); + if ((ret = + __os_malloc(dbc->dbp->pgsize, NULL, &new_curs->split_buf)) != 0) { + __os_free(new_curs, sizeof(*new_curs)); return (ret); - - DB_THREAD_LOCK(dbp); - TAILQ_INSERT_TAIL(&dbp->curs_queue, *dbcp, links); - DB_THREAD_UNLOCK(dbp); - return (ret); -} - -static int -__ham_c_init(dbp, txnid, dbcp) - DB *dbp; - DB_TXN *txnid; - DBC **dbcp; -{ - DBC *db_curs; - HASH_CURSOR *new_curs; - - if ((db_curs = (DBC *)__db_calloc(sizeof(DBC), 1)) == NULL) - return (ENOMEM); - - if ((new_curs = - (HASH_CURSOR *)__db_calloc(sizeof(struct cursor_t), 1)) == NULL) { - FREE(db_curs, sizeof(DBC)); - return (ENOMEM); } - db_curs->internal = new_curs; - db_curs->c_close = __ham_c_close; - db_curs->c_del = __ham_c_del; - db_curs->c_get = __ham_c_get; - db_curs->c_put = __ham_c_put; - db_curs->txn = txnid; - db_curs->dbp = dbp; + new_curs->dbc = dbc; + + dbc->internal = new_curs; + dbc->c_am_close = __ham_c_close; + dbc->c_am_destroy = __ham_c_destroy; + dbc->c_del = __ham_c_del; + dbc->c_get = __ham_c_get; + dbc->c_put = __ham_c_put; - new_curs->db_cursor = db_curs; __ham_item_init(new_curs); - if (dbcp != NULL) - *dbcp = db_curs; return (0); } +/* + * __ham_c_close -- + * Close down the cursor from a single use. + */ static int -__ham_delete(dbp, txn, key, flags) - DB *dbp; - DB_TXN *txn; - DBT *key; - u_int32_t flags; -{ - DB *ldbp; - HTAB *hashp; - HASH_CURSOR *hcp; - int ret, t_ret; - - DEBUG_LWRITE(dbp, txn, "ham_delete", key, NULL, flags); - if ((ret = - __db_delchk(dbp, key, flags, F_ISSET(dbp, DB_AM_RDONLY))) != 0) - return (ret); - - ldbp = dbp; - if (F_ISSET(dbp, DB_AM_THREAD) && - (ret = __db_gethandle(dbp, __ham_hdup, &ldbp)) != 0) - return (ret); - hashp = (HTAB *)ldbp->internal; - SET_LOCKER(ldbp, txn); - GET_META(ldbp, hashp); - hcp = TAILQ_FIRST(&ldbp->curs_queue)->internal; - - hashp->hash_accesses++; - if ((ret = __ham_lookup(hashp, hcp, key, 0, DB_LOCK_WRITE)) == 0) { - if (F_ISSET(hcp, H_OK)) - ret = __ham_del_pair(hashp, hcp, 1); - else - ret = DB_NOTFOUND; - } - - if ((t_ret = __ham_item_done(hashp, hcp, ret == 0)) != 0 && ret == 0) - ret = t_ret; - RELEASE_META(ldbp, hashp); - if (F_ISSET(dbp, DB_AM_THREAD)) - __db_puthandle(ldbp); - return (ret); -} - -/* ****************** CURSORS ********************************** */ -static int -__ham_c_close(cursor) - DBC *cursor; +__ham_c_close(dbc) + DBC *dbc; { - DB *ldbp; int ret; - DEBUG_LWRITE(cursor->dbp, cursor->txn, "ham_c_close", NULL, NULL, 0); - /* - * If the pagep, dpagep, and lock fields of the cursor are all NULL, - * then there really isn't a need to get a handle here. However, - * the normal case is that at least one of those fields is non-NULL, - * and putting those checks in here would couple the ham_item_done - * functionality with cursor close which would be pretty disgusting. - * Instead, we pay the overhead here of always getting the handle. - */ - ldbp = cursor->dbp; - if (F_ISSET(cursor->dbp, DB_AM_THREAD) && - (ret = __db_gethandle(cursor->dbp, __ham_hdup, &ldbp)) != 0) + if ((ret = __ham_item_done(dbc, 0)) != 0) return (ret); - ret = __ham_c_iclose(ldbp, cursor); - - if (F_ISSET(ldbp, DB_AM_THREAD)) - __db_puthandle(ldbp); - return (ret); + __ham_item_init((HASH_CURSOR *)dbc->internal); + return (0); } + /* - * __ham_c_iclose -- - * - * Internal cursor close routine; assumes it is being passed the correct - * handle, rather than getting and putting a handle. - * - * PUBLIC: int __ham_c_iclose __P((DB *, DBC *)); + * __ham_c_destroy -- + * Cleanup the access method private part of a cursor. */ -int -__ham_c_iclose(dbp, dbc) - DB *dbp; +static int +__ham_c_destroy(dbc) DBC *dbc; { HASH_CURSOR *hcp; - HTAB *hashp; - int ret; - hashp = (HTAB *)dbp->internal; hcp = (HASH_CURSOR *)dbc->internal; - ret = __ham_item_done(hashp, hcp, 0); - - if (hcp->big_key) - FREE(hcp->big_key, hcp->big_keylen); - if (hcp->big_data) - FREE(hcp->big_data, hcp->big_datalen); + if (hcp->split_buf != NULL) + __os_free(hcp->split_buf, dbc->dbp->pgsize); + __os_free(hcp, sizeof(HASH_CURSOR)); - /* - * All cursors (except the default ones) are linked off the master. - * Therefore, when we close the cursor, we have to remove it from - * the master, not the local one. - * XXX I am always removing from the master; what about local cursors? - */ - DB_THREAD_LOCK(dbc->dbp); - TAILQ_REMOVE(&dbc->dbp->curs_queue, dbc, links); - DB_THREAD_UNLOCK(dbc->dbp); - - FREE(hcp, sizeof(HASH_CURSOR)); - FREE(dbc, sizeof(DBC)); - - return (ret); + return (0); } static int -__ham_c_del(cursor, flags) - DBC *cursor; +__ham_c_del(dbc, flags) + DBC *dbc; u_int32_t flags; { - DB *ldbp; + DB *dbp; + DBT repldbt; HASH_CURSOR *hcp; HASH_CURSOR save_curs; - HTAB *hashp; db_pgno_t ppgno, chg_pgno; int ret, t_ret; - DEBUG_LWRITE(cursor->dbp, cursor->txn, "ham_c_del", NULL, NULL, flags); - ldbp = cursor->dbp; - if (F_ISSET(cursor->dbp, DB_AM_THREAD) && - (ret = __db_gethandle(cursor->dbp, __ham_hdup, &ldbp)) != 0) - return (ret); - hashp = (HTAB *)ldbp->internal; - hcp = (HASH_CURSOR *)cursor->internal; - save_curs = *hcp; - if ((ret = __db_cdelchk(ldbp, flags, - F_ISSET(ldbp, DB_AM_RDONLY), IS_VALID(hcp))) != 0) + DEBUG_LWRITE(dbc, dbc->txn, "ham_c_del", NULL, NULL, flags); + dbp = dbc->dbp; + DB_PANIC_CHECK(dbp); + hcp = (HASH_CURSOR *)dbc->internal; + + if ((ret = __db_cdelchk(dbc->dbp, flags, + F_ISSET(dbc->dbp, DB_AM_RDONLY), IS_VALID(hcp))) != 0) return (ret); + if (F_ISSET(hcp, H_DELETED)) return (DB_NOTFOUND); - SET_LOCKER(hashp->dbp, cursor->txn); - GET_META(hashp->dbp, hashp); - hashp->hash_accesses++; - if ((ret = __ham_get_cpage(hashp, hcp, DB_LOCK_WRITE)) != 0) + /* + * If we are in the concurrent DB product and this cursor + * is not a write cursor, then this request is invalid. + * If it is a simple write cursor, then we need to upgrade its + * lock. + */ + if (F_ISSET(dbp, DB_AM_CDB)) { + /* Make sure it's a valid update cursor. */ + if (!F_ISSET(dbc, DBC_RMW | DBC_WRITER)) + return (EINVAL); + + if (F_ISSET(dbc, DBC_RMW) && + (ret = lock_get(dbp->dbenv->lk_info, dbc->locker, + DB_LOCK_UPGRADE, &dbc->lock_dbt, DB_LOCK_WRITE, + &dbc->mylock)) != 0) + return (EAGAIN); + } + + GET_META(dbp, hcp, ret); + if (ret != 0) + return (ret); + + SAVE_CURSOR(hcp, &save_curs); + hcp->stats.hash_deleted++; + + if ((ret = __ham_get_cpage(dbc, DB_LOCK_WRITE)) != 0) goto out; if (F_ISSET(hcp, H_ISDUP) && hcp->dpgno != PGNO_INVALID) { /* @@ -695,20 +428,20 @@ __ham_c_del(cursor, flags) /* Remove item from duplicate page. */ chg_pgno = hcp->dpgno; - if ((ret = __db_drem(hashp->dbp, + if ((ret = __db_drem(dbc, &hcp->dpagep, hcp->dndx, __ham_del_page)) != 0) goto out; if (hcp->dpagep == NULL) { if (ppgno != PGNO_INVALID) { /* Case 3 */ hcp->dpgno = ppgno; - if ((ret = __ham_get_cpage(hashp, hcp, + if ((ret = __ham_get_cpage(dbc, DB_LOCK_READ)) != 0) goto out; hcp->dndx = NUM_ENT(hcp->dpagep); F_SET(hcp, H_DELETED); } else { /* Case 4 */ - ret = __ham_del_pair(hashp, hcp, 1); + ret = __ham_del_pair(dbc, 1); hcp->dpgno = PGNO_INVALID; /* * Delpair updated the cursor queue, so we @@ -723,6 +456,15 @@ __ham_c_del(cursor, flags) memcpy(HOFFDUP_PGNO(P_ENTRY(hcp->pagep, H_DATAINDEX(hcp->bndx))), &hcp->dpgno, sizeof(db_pgno_t)); + /* + * We need to put the master page here, because + * although we have a duplicate page, the master + * page is dirty, and ham_item_done assumes that + * if you have a duplicate page, it's the only one + * that can be dirty. + */ + ret = __ham_put_page(dbp, hcp->pagep, 1); + hcp->pagep = NULL; F_SET(hcp, H_DELETED); } else /* Case 1 */ F_SET(hcp, H_DELETED); @@ -730,17 +472,17 @@ __ham_c_del(cursor, flags) __ham_c_update(hcp, chg_pgno, 0, 0, 1); } else if (F_ISSET(hcp, H_ISDUP)) { /* on page */ if (hcp->dup_off == 0 && DUP_SIZE(hcp->dup_len) == - LEN_HDATA(hcp->pagep, hashp->hdr->pagesize, hcp->bndx)) - ret = __ham_del_pair(hashp, hcp, 1); + LEN_HDATA(hcp->pagep, hcp->hdr->pagesize, hcp->bndx)) + ret = __ham_del_pair(dbc, 1); else { - DBT repldbt; - repldbt.flags = 0; F_SET(&repldbt, DB_DBT_PARTIAL); repldbt.doff = hcp->dup_off; repldbt.dlen = DUP_SIZE(hcp->dup_len); repldbt.size = 0; - ret = __ham_replpair(hashp, hcp, &repldbt, 0); + repldbt.data = + HKEYDATA_DATA(H_PAIRDATA(hcp->pagep, hcp->bndx)); + ret = __ham_replpair(dbc, &repldbt, 0); hcp->dup_tlen -= DUP_SIZE(hcp->dup_len); F_SET(hcp, H_DELETED); __ham_c_update(hcp, hcp->pgno, @@ -749,48 +491,53 @@ __ham_c_del(cursor, flags) } else /* Not a duplicate */ -normal: ret = __ham_del_pair(hashp, hcp, 1); +normal: ret = __ham_del_pair(dbc, 1); -out: if ((t_ret = __ham_item_done(hashp, hcp, ret == 0)) != 0 && ret == 0) +out: if ((t_ret = __ham_item_done(dbc, ret == 0)) != 0 && ret == 0) ret = t_ret; - if (ret != 0) - *hcp = save_curs; - RELEASE_META(hashp->dbp, hashp); - if (F_ISSET(cursor->dbp, DB_AM_THREAD)) - __db_puthandle(ldbp); + RELEASE_META(dbp, hcp); + RESTORE_CURSOR(dbp, hcp, &save_curs, ret); + if (F_ISSET(dbp, DB_AM_CDB) && F_ISSET(dbc, DBC_RMW)) + (void)__lock_downgrade(dbp->dbenv->lk_info, dbc->mylock, + DB_LOCK_IWRITE, 0); return (ret); } static int -__ham_c_get(cursor, key, data, flags) - DBC *cursor; +__ham_c_get(dbc, key, data, flags) + DBC *dbc; DBT *key; DBT *data; u_int32_t flags; { - DB *ldbp; - HTAB *hashp; + DB *dbp; HASH_CURSOR *hcp, save_curs; + db_lockmode_t lock_type; int get_key, ret, t_ret; - DEBUG_LREAD(cursor->dbp, cursor->txn, "ham_c_get", + DEBUG_LREAD(dbc, dbc->txn, "ham_c_get", flags == DB_SET || flags == DB_SET_RANGE ? key : NULL, NULL, flags); - ldbp = cursor->dbp; - if (F_ISSET(cursor->dbp, DB_AM_THREAD) && - (ret = __db_gethandle(cursor->dbp, __ham_hdup, &ldbp)) != 0) - return (ret); - hashp = (HTAB *)(ldbp->internal); - hcp = (HASH_CURSOR *)cursor->internal; - save_curs = *hcp; + + hcp = (HASH_CURSOR *)dbc->internal; + dbp = dbc->dbp; + DB_PANIC_CHECK(dbp); + SAVE_CURSOR(hcp, &save_curs); if ((ret = - __db_cgetchk(hashp->dbp, key, data, flags, IS_VALID(hcp))) != 0) + __db_cgetchk(dbp, key, data, flags, IS_VALID(hcp))) != 0) return (ret); - SET_LOCKER(hashp->dbp, cursor->txn); - GET_META(hashp->dbp, hashp); - hashp->hash_accesses++; + /* Clear OR'd in additional bits so we can check for flag equality. */ + if (LF_ISSET(DB_RMW)) { + lock_type = DB_LOCK_WRITE; + LF_CLR(DB_RMW); + } else + lock_type = DB_LOCK_READ; + GET_META(dbp, hcp, ret); + if (ret != 0) + return (ret); + hcp->stats.hash_get++; hcp->seek_size = 0; ret = 0; @@ -798,24 +545,39 @@ __ham_c_get(cursor, key, data, flags) switch (flags) { case DB_PREV: if (hcp->bucket != BUCKET_INVALID) { - ret = __ham_item_prev(hashp, hcp, DB_LOCK_READ); + ret = __ham_item_prev(dbc, lock_type); break; } /* FALLTHROUGH */ case DB_LAST: - ret = __ham_item_last(hashp, hcp, DB_LOCK_READ); + ret = __ham_item_last(dbc, lock_type); break; case DB_FIRST: - ret = __ham_item_first(hashp, hcp, DB_LOCK_READ); + ret = __ham_item_first(dbc, lock_type); + break; + case DB_NEXT_DUP: + if (hcp->bucket == BUCKET_INVALID) + ret = EINVAL; + else { + F_SET(hcp, H_DUPONLY); + ret = __ham_item_next(dbc, lock_type); + } break; case DB_NEXT: if (hcp->bucket == BUCKET_INVALID) hcp->bucket = 0; - ret = __ham_item_next(hashp, hcp, DB_LOCK_READ); + ret = __ham_item_next(dbc, lock_type); break; case DB_SET: case DB_SET_RANGE: - ret = __ham_lookup(hashp, hcp, key, 0, DB_LOCK_READ); + case DB_GET_BOTH: + if (F_ISSET(dbc, DBC_CONTINUE)) { + F_SET(hcp, H_DUPONLY); + ret = __ham_item_next(dbc, lock_type); + } else if (F_ISSET(dbc, DBC_KEYSET)) + ret = __ham_item(dbc, lock_type); + else + ret = __ham_lookup(dbc, key, 0, lock_type); get_key = 0; break; case DB_CURRENT: @@ -824,7 +586,7 @@ __ham_c_get(cursor, key, data, flags) goto out; } - ret = __ham_item(hashp, hcp, DB_LOCK_READ); + ret = __ham_item(dbc, lock_type); break; } @@ -837,12 +599,12 @@ __ham_c_get(cursor, key, data, flags) goto out1; else if (F_ISSET(hcp, H_OK)) { /* Get the key. */ - if (get_key && (ret = __db_ret(hashp->dbp, hcp->pagep, - H_KEYINDEX(hcp->bndx), key, &hcp->big_key, - &hcp->big_keylen)) != 0) + if (get_key && (ret = __db_ret(dbp, hcp->pagep, + H_KEYINDEX(hcp->bndx), key, &dbc->rkey.data, + &dbc->rkey.size)) != 0) goto out1; - ret = __ham_dup_return(hashp, hcp, data, flags); + ret = __ham_dup_return(dbc, data, flags); break; } else if (!F_ISSET(hcp, H_NOMORE)) { abort(); @@ -855,7 +617,7 @@ __ham_c_get(cursor, key, data, flags) switch (flags) { case DB_LAST: case DB_PREV: - ret = __ham_item_done(hashp, hcp, 0); + ret = __ham_item_done(dbc, 0); if (hcp->bucket == 0) { ret = DB_NOTFOUND; goto out1; @@ -863,24 +625,24 @@ __ham_c_get(cursor, key, data, flags) hcp->bucket--; hcp->bndx = NDX_INVALID; if (ret == 0) - ret = __ham_item_prev(hashp, - hcp, DB_LOCK_READ); + ret = __ham_item_prev(dbc, lock_type); break; case DB_FIRST: case DB_NEXT: - ret = __ham_item_done(hashp, hcp, 0); + ret = __ham_item_done(dbc, 0); hcp->bndx = NDX_INVALID; hcp->bucket++; hcp->pgno = PGNO_INVALID; hcp->pagep = NULL; - if (hcp->bucket > hashp->hdr->max_bucket) { + if (hcp->bucket > hcp->hdr->max_bucket) { ret = DB_NOTFOUND; goto out1; } if (ret == 0) - ret = __ham_item_next(hashp, - hcp, DB_LOCK_READ); + ret = __ham_item_next(dbc, lock_type); break; + case DB_GET_BOTH: + case DB_NEXT_DUP: case DB_SET: case DB_SET_RANGE: /* Key not found. */ @@ -888,85 +650,137 @@ __ham_c_get(cursor, key, data, flags) goto out1; } } -out1: if ((t_ret = __ham_item_done(hashp, hcp, 0)) != 0 && ret == 0) +out1: if ((t_ret = __ham_item_done(dbc, 0)) != 0 && ret == 0) ret = t_ret; -out: if (ret) - *hcp = save_curs; - RELEASE_META(hashp->dbp, hashp); - if (F_ISSET(cursor->dbp, DB_AM_THREAD)) - __db_puthandle(ldbp); +out: RELEASE_META(dbp, hcp); + RESTORE_CURSOR(dbp, hcp, &save_curs, ret); return (ret); } static int -__ham_c_put(cursor, key, data, flags) - DBC *cursor; +__ham_c_put(dbc, key, data, flags) + DBC *dbc; DBT *key; DBT *data; u_int32_t flags; { - DB *ldbp; + DB *dbp; + DBT tmp_val, *myval; HASH_CURSOR *hcp, save_curs; - HTAB *hashp; u_int32_t nbytes; int ret, t_ret; - DEBUG_LWRITE(cursor->dbp, cursor->txn, "ham_c_put", + dbp = dbc->dbp; + DB_PANIC_CHECK(dbp); + DEBUG_LWRITE(dbc, dbc->txn, "ham_c_put", flags == DB_KEYFIRST || flags == DB_KEYLAST ? key : NULL, data, flags); - ldbp = cursor->dbp; - if (F_ISSET(cursor->dbp, DB_AM_THREAD) && - (ret = __db_gethandle(cursor->dbp, __ham_hdup, &ldbp)) != 0) - return (ret); - hashp = (HTAB *)(ldbp->internal); - hcp = (HASH_CURSOR *)cursor->internal; - save_curs = *hcp; + hcp = (HASH_CURSOR *)dbc->internal; - if ((ret = __db_cputchk(hashp->dbp, key, data, flags, - F_ISSET(ldbp, DB_AM_RDONLY), IS_VALID(hcp))) != 0) + if ((ret = __db_cputchk(dbp, key, data, flags, + F_ISSET(dbp, DB_AM_RDONLY), IS_VALID(hcp))) != 0) return (ret); - if (F_ISSET(hcp, H_DELETED)) + + if (F_ISSET(hcp, H_DELETED) && + flags != DB_KEYFIRST && flags != DB_KEYLAST) return (DB_NOTFOUND); - SET_LOCKER(hashp->dbp, cursor->txn); - GET_META(hashp->dbp, hashp); - ret = 0; + /* + * If we are in the concurrent DB product and this cursor + * is not a write cursor, then this request is invalid. + * If it is a simple write cursor, then we need to upgrade its + * lock. + */ + if (F_ISSET(dbp, DB_AM_CDB)) { + /* Make sure it's a valid update cursor. */ + if (!F_ISSET(dbc, DBC_RMW | DBC_WRITER)) + return (EINVAL); + + if (F_ISSET(dbc, DBC_RMW) && + (ret = lock_get(dbp->dbenv->lk_info, dbc->locker, + DB_LOCK_UPGRADE, &dbc->lock_dbt, DB_LOCK_WRITE, + &dbc->mylock)) != 0) + return (EAGAIN); + } + + GET_META(dbp, hcp, ret); + if (ret != 0) + return (ret); + + SAVE_CURSOR(hcp, &save_curs); + hcp->stats.hash_put++; switch (flags) { case DB_KEYLAST: case DB_KEYFIRST: - nbytes = (ISBIG(hashp, key->size) ? HOFFPAGE_PSIZE : + nbytes = (ISBIG(hcp, key->size) ? HOFFPAGE_PSIZE : HKEYDATA_PSIZE(key->size)) + - (ISBIG(hashp, data->size) ? HOFFPAGE_PSIZE : + (ISBIG(hcp, data->size) ? HOFFPAGE_PSIZE : HKEYDATA_PSIZE(data->size)); - ret = __ham_lookup(hashp, hcp, key, nbytes, DB_LOCK_WRITE); + if ((ret = __ham_lookup(dbc, + key, nbytes, DB_LOCK_WRITE)) == DB_NOTFOUND) { + ret = 0; + if (hcp->seek_found_page != PGNO_INVALID && + hcp->seek_found_page != hcp->pgno) { + if ((ret = __ham_item_done(dbc, 0)) != 0) + goto out; + hcp->pgno = hcp->seek_found_page; + hcp->bndx = NDX_INVALID; + } + + if (F_ISSET(data, DB_DBT_PARTIAL) && data->doff != 0) { + /* + * A partial put, but the key does not exist + * and we are not beginning the write at 0. + * We must create a data item padded up to doff + * and then write the new bytes represented by + * val. + */ + if ((ret = __ham_init_dbt(&tmp_val, + data->size + data->doff, + &dbc->rdata.data, &dbc->rdata.size)) == 0) { + memset(tmp_val.data, 0, data->doff); + memcpy((u_int8_t *)tmp_val.data + + data->doff, data->data, data->size); + myval = &tmp_val; + } + } else + myval = (DBT *)data; + + if (ret == 0) + ret = __ham_add_el(dbc, key, myval, H_KEYDATA); + goto done; + } break; case DB_BEFORE: case DB_AFTER: case DB_CURRENT: - ret = __ham_item(hashp, hcp, DB_LOCK_WRITE); + ret = __ham_item(dbc, DB_LOCK_WRITE); break; } if (ret == 0) { - if (flags == DB_CURRENT && !F_ISSET(ldbp, DB_AM_DUP)) - ret = __ham_overwrite(hashp, hcp, data); + if ((flags == DB_CURRENT && !F_ISSET(hcp, H_ISDUP)) || + ((flags == DB_KEYFIRST || flags == DB_KEYLAST) && + !F_ISSET(dbp, DB_AM_DUP))) + ret = __ham_overwrite(dbc, data); else - ret = __ham_add_dup(hashp, hcp, data, flags); + ret = __ham_add_dup(dbc, data, flags); } - if (ret == 0 && F_ISSET(hcp, H_EXPAND)) { - ret = __ham_expand_table(hashp); +done: if (ret == 0 && F_ISSET(hcp, H_EXPAND)) { + ret = __ham_expand_table(dbc); F_CLR(hcp, H_EXPAND); } - if ((t_ret = __ham_item_done(hashp, hcp, ret == 0)) != 0 && ret == 0) + if ((t_ret = __ham_item_done(dbc, ret == 0)) != 0 && ret == 0) ret = t_ret; - if (ret != 0) - *hcp = save_curs; - RELEASE_META(hashp->dbp, hashp); - if (F_ISSET(cursor->dbp, DB_AM_THREAD)) - __db_puthandle(ldbp); + +out: RELEASE_META(dbp, hcp); + RESTORE_CURSOR(dbp, hcp, &save_curs, ret); + if (F_ISSET(dbp, DB_AM_CDB) && F_ISSET(dbc, DBC_RMW)) + (void)__lock_downgrade(dbp->dbenv->lk_info, dbc->mylock, + DB_LOCK_IWRITE, 0); return (ret); } @@ -974,19 +788,21 @@ __ham_c_put(cursor, key, data, flags) /* * __ham_expand_table -- - * - * PUBLIC: int __ham_expand_table __P((HTAB *)); */ -int -__ham_expand_table(hashp) - HTAB *hashp; +static int +__ham_expand_table(dbc) + DBC *dbc; { + DB *dbp; + HASH_CURSOR *hcp; DB_LSN new_lsn; u_int32_t old_bucket, new_bucket, spare_ndx; int ret; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; ret = 0; - DIRTY_META(hashp, ret); + DIRTY_META(dbp, hcp, ret); if (ret) return (ret); @@ -999,78 +815,78 @@ __ham_expand_table(hashp) * see what the log of one greater than that is; here we have to * look at the log of max + 2. VERY NASTY STUFF. */ - if (__db_log2(hashp->hdr->max_bucket + 2) > hashp->hdr->ovfl_point) { + if (__db_log2(hcp->hdr->max_bucket + 2) > hcp->hdr->ovfl_point) { /* * We are about to shift the split point. Make sure that * if the next doubling is going to be big (more than 8 * pages), we have some extra pages around. */ - if (hashp->hdr->max_bucket + 1 >= 8 && - hashp->hdr->spares[hashp->hdr->ovfl_point] < - hashp->hdr->spares[hashp->hdr->ovfl_point - 1] + - hashp->hdr->ovfl_point + 1) - __ham_init_ovflpages(hashp); + if (hcp->hdr->max_bucket + 1 >= 8 && + hcp->hdr->spares[hcp->hdr->ovfl_point] < + hcp->hdr->spares[hcp->hdr->ovfl_point - 1] + + hcp->hdr->ovfl_point + 1) + __ham_init_ovflpages(dbc); } /* Now we can log the meta-data split. */ - if (DB_LOGGING(hashp->dbp)) { - if ((ret = __ham_splitmeta_log(hashp->dbp->dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, - hashp->dbp->log_fileid, - hashp->hdr->max_bucket, hashp->hdr->ovfl_point, - hashp->hdr->spares[hashp->hdr->ovfl_point], - &hashp->hdr->lsn)) != 0) + if (DB_LOGGING(dbc)) { + if ((ret = __ham_splitmeta_log(dbp->dbenv->lg_info, + dbc->txn, &new_lsn, 0, dbp->log_fileid, + hcp->hdr->max_bucket, hcp->hdr->ovfl_point, + hcp->hdr->spares[hcp->hdr->ovfl_point], + &hcp->hdr->lsn)) != 0) return (ret); - hashp->hdr->lsn = new_lsn; + hcp->hdr->lsn = new_lsn; } - hashp->hash_expansions++; - new_bucket = ++hashp->hdr->max_bucket; - old_bucket = (hashp->hdr->max_bucket & hashp->hdr->low_mask); + hcp->stats.hash_expansions++; + new_bucket = ++hcp->hdr->max_bucket; + old_bucket = (hcp->hdr->max_bucket & hcp->hdr->low_mask); /* * If the split point is increasing, copy the current contents * of the spare split bucket to the next bucket. */ - spare_ndx = __db_log2(hashp->hdr->max_bucket + 1); - if (spare_ndx > hashp->hdr->ovfl_point) { - hashp->hdr->spares[spare_ndx] = - hashp->hdr->spares[hashp->hdr->ovfl_point]; - hashp->hdr->ovfl_point = spare_ndx; + spare_ndx = __db_log2(hcp->hdr->max_bucket + 1); + if (spare_ndx > hcp->hdr->ovfl_point) { + hcp->hdr->spares[spare_ndx] = + hcp->hdr->spares[hcp->hdr->ovfl_point]; + hcp->hdr->ovfl_point = spare_ndx; } - if (new_bucket > hashp->hdr->high_mask) { + if (new_bucket > hcp->hdr->high_mask) { /* Starting a new doubling */ - hashp->hdr->low_mask = hashp->hdr->high_mask; - hashp->hdr->high_mask = new_bucket | hashp->hdr->low_mask; + hcp->hdr->low_mask = hcp->hdr->high_mask; + hcp->hdr->high_mask = new_bucket | hcp->hdr->low_mask; } - if (BUCKET_TO_PAGE(hashp, new_bucket) > MAX_PAGES(hashp)) { - __db_err(hashp->dbp->dbenv, + if (BUCKET_TO_PAGE(hcp, new_bucket) > MAX_PAGES(hcp)) { + __db_err(dbp->dbenv, "hash: Cannot allocate new bucket. Pages exhausted."); return (ENOSPC); } /* Relocate records to the new bucket */ - return (__ham_split_page(hashp, old_bucket, new_bucket)); + return (__ham_split_page(dbc, old_bucket, new_bucket)); } /* - * PUBLIC: u_int32_t __ham_call_hash __P((HTAB *, u_int8_t *, int32_t)); + * PUBLIC: u_int32_t __ham_call_hash __P((HASH_CURSOR *, u_int8_t *, int32_t)); */ u_int32_t -__ham_call_hash(hashp, k, len) - HTAB *hashp; +__ham_call_hash(hcp, k, len) + HASH_CURSOR *hcp; u_int8_t *k; int32_t len; { u_int32_t n, bucket; - n = (u_int32_t)hashp->hash(k, len); - bucket = n & hashp->hdr->high_mask; - if (bucket > hashp->hdr->max_bucket) - bucket = bucket & hashp->hdr->low_mask; + n = (u_int32_t)(hcp->dbc->dbp->h_hash(k, len)); + + bucket = n & hcp->hdr->high_mask; + if (bucket > hcp->hdr->max_bucket) + bucket = bucket & hcp->hdr->low_mask; return (bucket); } @@ -1079,31 +895,36 @@ __ham_call_hash(hashp, k, len) * everything held by the cursor. */ static int -__ham_dup_return(hashp, hcp, val, flags) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_dup_return(dbc, val, flags) + DBC *dbc; DBT *val; u_int32_t flags; { + DB *dbp; + HASH_CURSOR *hcp; PAGE *pp; DBT *myval, tmp_val; db_indx_t ndx; db_pgno_t pgno; + u_int32_t off, tlen; u_int8_t *hk, type; - int ret; + int cmp, ret; db_indx_t len; /* Check for duplicate and return the first one. */ + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; ndx = H_DATAINDEX(hcp->bndx); type = HPAGE_TYPE(hcp->pagep, ndx); pp = hcp->pagep; myval = val; /* - * There are 3 cases: + * There are 4 cases: * 1. We are not in duplicate, simply call db_ret. * 2. We are looking at keys and stumbled onto a duplicate. * 3. We are in the middle of a duplicate set. (ISDUP set) + * 4. This is a duplicate and we need to return a specific item. */ /* @@ -1115,7 +936,7 @@ __ham_dup_return(hashp, hcp, val, flags) if (type == H_DUPLICATE) { F_SET(hcp, H_ISDUP); hcp->dup_tlen = LEN_HDATA(hcp->pagep, - hashp->hdr->pagesize, hcp->bndx); + hcp->hdr->pagesize, hcp->bndx); hk = H_PAIRDATA(hcp->pagep, hcp->bndx); if (flags == DB_LAST || flags == DB_PREV) { hcp->dndx = 0; @@ -1141,18 +962,63 @@ __ham_dup_return(hashp, hcp, val, flags) memcpy(&pgno, HOFFDUP_PGNO(P_ENTRY(hcp->pagep, ndx)), sizeof(db_pgno_t)); if (flags == DB_LAST || flags == DB_PREV) { - if ((ret = __db_dend(hashp->dbp, + if ((ret = __db_dend(dbc, pgno, &hcp->dpagep)) != 0) return (ret); hcp->dpgno = PGNO(hcp->dpagep); hcp->dndx = NUM_ENT(hcp->dpagep) - 1; - } else if ((ret = __ham_next_cpage(hashp, - hcp, pgno, 0, H_ISDUP)) != 0) + } else if ((ret = __ham_next_cpage(dbc, + pgno, 0, H_ISDUP)) != 0) return (ret); } } /* + * If we are retrieving a specific key/data pair, then we + * may need to adjust the cursor before returning data. + */ + if (flags == DB_GET_BOTH) { + if (F_ISSET(hcp, H_ISDUP)) { + if (hcp->dpgno != PGNO_INVALID) { + if ((ret = __db_dsearch(dbc, 0, val, + hcp->dpgno, &hcp->dndx, &hcp->dpagep, &cmp)) + != 0) + return (ret); + if (cmp == 0) + hcp->dpgno = PGNO(hcp->dpagep); + } else { + __ham_dsearch(dbc, val, &off, &cmp); + hcp->dup_off = off; + } + } else { + hk = H_PAIRDATA(hcp->pagep, hcp->bndx); + if (((HKEYDATA *)hk)->type == H_OFFPAGE) { + memcpy(&tlen, + HOFFPAGE_TLEN(hk), sizeof(u_int32_t)); + memcpy(&pgno, + HOFFPAGE_PGNO(hk), sizeof(db_pgno_t)); + if ((ret = __db_moff(dbp, val, + pgno, tlen, dbp->dup_compare, &cmp)) != 0) + return (ret); + } else { + /* + * We do not zero tmp_val since the comparison + * routines may only look at data and size. + */ + tmp_val.data = HKEYDATA_DATA(hk); + tmp_val.size = LEN_HDATA(hcp->pagep, + dbp->pgsize, hcp->bndx); + cmp = dbp->dup_compare == NULL ? + __bam_defcmp(&tmp_val, val) : + dbp->dup_compare(&tmp_val, val); + } + } + + if (cmp != 0) + return (DB_NOTFOUND); + } + + /* * Now, everything is initialized, grab a duplicate if * necessary. */ @@ -1162,14 +1028,34 @@ __ham_dup_return(hashp, hcp, val, flags) ndx = hcp->dndx; } else { /* - * Copy the DBT in case we are retrieving into - * user memory and we need the parameters for - * it. + * Copy the DBT in case we are retrieving into user + * memory and we need the parameters for it. If the + * user requested a partial, then we need to adjust + * the user's parameters to get the partial of the + * duplicate which is itself a partial. */ memcpy(&tmp_val, val, sizeof(*val)); - F_SET(&tmp_val, DB_DBT_PARTIAL); - tmp_val.dlen = hcp->dup_len; - tmp_val.doff = hcp->dup_off + sizeof(db_indx_t); + if (F_ISSET(&tmp_val, DB_DBT_PARTIAL)) { + /* + * Take the user's length unless it would go + * beyond the end of the duplicate. + */ + if (tmp_val.doff + hcp->dup_off > hcp->dup_len) + tmp_val.dlen = 0; + else if (tmp_val.dlen + tmp_val.doff > + hcp->dup_len) + tmp_val.dlen = + hcp->dup_len - tmp_val.doff; + + /* + * Calculate the new offset. + */ + tmp_val.doff += hcp->dup_off; + } else { + F_SET(&tmp_val, DB_DBT_PARTIAL); + tmp_val.dlen = hcp->dup_len; + tmp_val.doff = hcp->dup_off + sizeof(db_indx_t); + } myval = &tmp_val; } } @@ -1178,8 +1064,8 @@ __ham_dup_return(hashp, hcp, val, flags) * Finally, if we had a duplicate, pp, ndx, and myval should be * set appropriately. */ - if ((ret = __db_ret(hashp->dbp, pp, ndx, myval, &hcp->big_data, - &hcp->big_datalen)) != 0) + if ((ret = __db_ret(dbp, pp, ndx, myval, &dbc->rdata.data, + &dbc->rdata.size)) != 0) return (ret); /* @@ -1193,16 +1079,17 @@ __ham_dup_return(hashp, hcp, val, flags) } static int -__ham_overwrite(hashp, hcp, nval) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_overwrite(dbc, nval) + DBC *dbc; DBT *nval; { + HASH_CURSOR *hcp; DBT *myval, tmp_val; u_int8_t *hk; - if (F_ISSET(hashp->dbp, DB_AM_DUP)) - return (__ham_add_dup(hashp, hcp, nval, DB_KEYLAST)); + hcp = (HASH_CURSOR *)dbc->internal; + if (F_ISSET(dbc->dbp, DB_AM_DUP)) + return (__ham_add_dup(dbc, nval, DB_KEYLAST)); else if (!F_ISSET(nval, DB_DBT_PARTIAL)) { /* Put/overwrite */ memcpy(&tmp_val, nval, sizeof(*nval)); @@ -1214,12 +1101,12 @@ __ham_overwrite(hashp, hcp, nval) HOFFPAGE_TLEN(hk), sizeof(u_int32_t)); else tmp_val.dlen = LEN_HDATA(hcp->pagep, - hashp->hdr->pagesize,hcp->bndx); + hcp->hdr->pagesize,hcp->bndx); myval = &tmp_val; } else /* Regular partial put */ myval = nval; - return (__ham_replpair(hashp, hcp, myval, 0)); + return (__ham_replpair(dbc, myval, 0)); } /* @@ -1232,29 +1119,32 @@ __ham_overwrite(hashp, hcp, nval) * non of the cursor pointer field are valid. */ static int -__ham_lookup(hashp, hcp, key, sought, mode) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_lookup(dbc, key, sought, mode) + DBC *dbc; const DBT *key; u_int32_t sought; db_lockmode_t mode; { + DB *dbp; + HASH_CURSOR *hcp; db_pgno_t pgno; u_int32_t tlen; int match, ret, t_ret; u_int8_t *hk; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; /* * Set up cursor so that we're looking for space to add an item * as we cycle through the pages looking for the key. */ - if ((ret = __ham_item_reset(hashp, hcp)) != 0) + if ((ret = __ham_item_reset(dbc)) != 0) return (ret); hcp->seek_size = sought; - hcp->bucket = __ham_call_hash(hashp, (u_int8_t *)key->data, key->size); + hcp->bucket = __ham_call_hash(hcp, (u_int8_t *)key->data, key->size); while (1) { - if ((ret = __ham_item_next(hashp, hcp, mode)) != 0) + if ((ret = __ham_item_next(dbc, mode)) != 0) return (ret); if (F_ISSET(hcp, H_NOMORE)) @@ -1267,7 +1157,9 @@ __ham_lookup(hashp, hcp, key, sought, mode) if (tlen == key->size) { memcpy(&pgno, HOFFPAGE_PGNO(hk), sizeof(db_pgno_t)); - match = __db_moff(hashp->dbp, key, pgno); + if ((ret = __db_moff(dbp, + key, pgno, tlen, NULL, &match)) != 0) + return (ret); if (match == 0) { F_SET(hcp, H_OK); return (0); @@ -1276,7 +1168,7 @@ __ham_lookup(hashp, hcp, key, sought, mode) break; case H_KEYDATA: if (key->size == LEN_HKEY(hcp->pagep, - hashp->hdr->pagesize, hcp->bndx) && + hcp->hdr->pagesize, hcp->bndx) && memcmp(key->data, HKEYDATA_DATA(hk), key->size) == 0) { F_SET(hcp, H_OK); @@ -1289,9 +1181,9 @@ __ham_lookup(hashp, hcp, key, sought, mode) * These are errors because keys are never * duplicated, only data items are. */ - return (__db_pgfmt(hashp->dbp, PGNO(hcp->pagep))); + return (__db_pgfmt(dbp, PGNO(hcp->pagep))); } - hashp->hash_collisions++; + hcp->stats.hash_collisions++; } /* @@ -1301,7 +1193,7 @@ __ham_lookup(hashp, hcp, key, sought, mode) if (sought != 0) return (ret); - if ((t_ret = __ham_item_done(hashp, hcp, 0)) != 0 && ret == 0) + if ((t_ret = __ham_item_done(dbc, 0)) != 0 && ret == 0) ret = t_ret; return (ret); } @@ -1318,12 +1210,13 @@ __ham_init_dbt(dbt, size, bufp, sizep) void **bufp; u_int32_t *sizep; { + int ret; + memset(dbt, 0, sizeof(*dbt)); if (*sizep < size) { - if ((*bufp = (void *)(*bufp == NULL ? - __db_malloc(size) : __db_realloc(*bufp, size))) == NULL) { + if ((ret = __os_realloc(bufp, size)) != 0) { *sizep = 0; - return (ENOMEM); + return (ret); } *sizep = size; } @@ -1352,8 +1245,8 @@ __ham_c_update(hcp, chg_pgno, len, add, is_dup) u_int32_t len; int add, is_dup; { + DB *dbp; DBC *cp; - HTAB *hp; HASH_CURSOR *lcp; int page_deleted; @@ -1379,10 +1272,10 @@ __ham_c_update(hcp, chg_pgno, len, add, is_dup) page_deleted = chg_pgno != PGNO_INVALID && chg_pgno != hcp->dpgno; - hp = hcp->db_cursor->dbp->master->internal; - DB_THREAD_LOCK(hp->dbp); + dbp = hcp->dbc->dbp; + DB_THREAD_LOCK(dbp); - for (cp = TAILQ_FIRST(&hp->dbp->curs_queue); cp != NULL; + for (cp = TAILQ_FIRST(&dbp->active_queue); cp != NULL; cp = TAILQ_NEXT(cp, links)) { if (cp->internal == hcp) continue; @@ -1440,43 +1333,5 @@ __ham_c_update(hcp, chg_pgno, len, add, is_dup) } } } - DB_THREAD_UNLOCK(hp->dbp); -} - -/* - * __ham_hdup -- - * This function gets called when we create a duplicate handle for a - * threaded DB. It should create the private part of the DB structure. - * - * PUBLIC: int __ham_hdup __P((DB *, DB *)); - */ -int -__ham_hdup(orig, new) - DB *orig, *new; -{ - DBC *curs; - HTAB *hashp; - int ret; - - if ((hashp = (HTAB *)__db_malloc(sizeof(HTAB))) == NULL) - return (ENOMEM); - - new->internal = hashp; - - hashp->dbp = new; - hashp->hlock = 0; - hashp->hdr = NULL; - hashp->hash = ((HTAB *)orig->internal)->hash; - if ((hashp->split_buf = (PAGE *)__db_malloc(orig->pgsize)) == NULL) - return (ENOMEM); - hashp->local_errno = 0; - hashp->hash_accesses = 0; - hashp->hash_collisions = 0; - hashp->hash_expansions = 0; - hashp->hash_overflows = 0; - hashp->hash_bigpages = 0; - /* Initialize the cursor queue. */ - ret = __ham_c_init(new, NULL, &curs); - TAILQ_INSERT_TAIL(&new->curs_queue, curs, links); - return (ret); + DB_THREAD_UNLOCK(dbp); } diff --git a/db2/hash/hash_auto.c b/db2/hash/hash_auto.c index 41b1ebed01..94a1dff6ed 100644 --- a/db2/hash/hash_auto.c +++ b/db2/hash/hash_auto.c @@ -10,7 +10,6 @@ #endif #include "db_int.h" -#include "shqueue.h" #include "db_page.h" #include "db_dispatch.h" #include "hash.h" @@ -46,8 +45,7 @@ int __ham_insdel_log(logp, txnid, ret_lsnp, flags, rectype = DB_ham_insdel; txn_num = txnid == NULL ? 0 : txnid->txnid; if (txnid == NULL) { - null_lsn.file = 0; - null_lsn.offset = 0; + ZERO_LSN(null_lsn); lsnp = &null_lsn; } else lsnp = &txnid->last_lsn; @@ -59,8 +57,8 @@ int __ham_insdel_log(logp, txnid, ret_lsnp, flags, + sizeof(*pagelsn) + sizeof(u_int32_t) + (key == NULL ? 0 : key->size) + sizeof(u_int32_t) + (data == NULL ? 0 : data->size); - if ((logrec.data = (void *)__db_malloc(logrec.size)) == NULL) - return (ENOMEM); + if ((ret = __os_malloc(logrec.size, NULL, &logrec.data)) != 0) + return (ret); bp = logrec.data; memcpy(bp, &rectype, sizeof(rectype)); @@ -109,7 +107,7 @@ int __ham_insdel_log(logp, txnid, ret_lsnp, flags, ret = log_put(logp, ret_lsnp, (DBT *)&logrec, flags); if (txnid != NULL) txnid->last_lsn = *ret_lsnp; - __db_free(logrec.data); + __os_free(logrec.data, 0); return (ret); } @@ -170,7 +168,7 @@ __ham_insdel_print(notused1, dbtp, lsnp, notused2, notused3) } printf("\n"); printf("\n"); - __db_free(argp); + __os_free(argp, 0); return (0); } @@ -184,11 +182,12 @@ __ham_insdel_read(recbuf, argpp) { __ham_insdel_args *argp; u_int8_t *bp; + int ret; - argp = (__ham_insdel_args *)__db_malloc(sizeof(__ham_insdel_args) + - sizeof(DB_TXN)); - if (argp == NULL) - return (ENOMEM); + ret = __os_malloc(sizeof(__ham_insdel_args) + + sizeof(DB_TXN), NULL, &argp); + if (ret != 0) + return (ret); argp->txnid = (DB_TXN *)&argp[1]; bp = recbuf; memcpy(&argp->type, bp, sizeof(argp->type)); @@ -250,8 +249,7 @@ int __ham_newpage_log(logp, txnid, ret_lsnp, flags, rectype = DB_ham_newpage; txn_num = txnid == NULL ? 0 : txnid->txnid; if (txnid == NULL) { - null_lsn.file = 0; - null_lsn.offset = 0; + ZERO_LSN(null_lsn); lsnp = &null_lsn; } else lsnp = &txnid->last_lsn; @@ -264,8 +262,8 @@ int __ham_newpage_log(logp, txnid, ret_lsnp, flags, + sizeof(*pagelsn) + sizeof(next_pgno) + sizeof(*nextlsn); - if ((logrec.data = (void *)__db_malloc(logrec.size)) == NULL) - return (ENOMEM); + if ((ret = __os_malloc(logrec.size, NULL, &logrec.data)) != 0) + return (ret); bp = logrec.data; memcpy(bp, &rectype, sizeof(rectype)); @@ -306,7 +304,7 @@ int __ham_newpage_log(logp, txnid, ret_lsnp, flags, ret = log_put(logp, ret_lsnp, (DBT *)&logrec, flags); if (txnid != NULL) txnid->last_lsn = *ret_lsnp; - __db_free(logrec.data); + __os_free(logrec.data, 0); return (ret); } @@ -354,7 +352,7 @@ __ham_newpage_print(notused1, dbtp, lsnp, notused2, notused3) printf("\tnextlsn: [%lu][%lu]\n", (u_long)argp->nextlsn.file, (u_long)argp->nextlsn.offset); printf("\n"); - __db_free(argp); + __os_free(argp, 0); return (0); } @@ -368,11 +366,12 @@ __ham_newpage_read(recbuf, argpp) { __ham_newpage_args *argp; u_int8_t *bp; + int ret; - argp = (__ham_newpage_args *)__db_malloc(sizeof(__ham_newpage_args) + - sizeof(DB_TXN)); - if (argp == NULL) - return (ENOMEM); + ret = __os_malloc(sizeof(__ham_newpage_args) + + sizeof(DB_TXN), NULL, &argp); + if (ret != 0) + return (ret); argp->txnid = (DB_TXN *)&argp[1]; bp = recbuf; memcpy(&argp->type, bp, sizeof(argp->type)); @@ -428,8 +427,7 @@ int __ham_splitmeta_log(logp, txnid, ret_lsnp, flags, rectype = DB_ham_splitmeta; txn_num = txnid == NULL ? 0 : txnid->txnid; if (txnid == NULL) { - null_lsn.file = 0; - null_lsn.offset = 0; + ZERO_LSN(null_lsn); lsnp = &null_lsn; } else lsnp = &txnid->last_lsn; @@ -439,8 +437,8 @@ int __ham_splitmeta_log(logp, txnid, ret_lsnp, flags, + sizeof(ovflpoint) + sizeof(spares) + sizeof(*metalsn); - if ((logrec.data = (void *)__db_malloc(logrec.size)) == NULL) - return (ENOMEM); + if ((ret = __os_malloc(logrec.size, NULL, &logrec.data)) != 0) + return (ret); bp = logrec.data; memcpy(bp, &rectype, sizeof(rectype)); @@ -469,7 +467,7 @@ int __ham_splitmeta_log(logp, txnid, ret_lsnp, flags, ret = log_put(logp, ret_lsnp, (DBT *)&logrec, flags); if (txnid != NULL) txnid->last_lsn = *ret_lsnp; - __db_free(logrec.data); + __os_free(logrec.data, 0); return (ret); } @@ -512,7 +510,7 @@ __ham_splitmeta_print(notused1, dbtp, lsnp, notused2, notused3) printf("\tmetalsn: [%lu][%lu]\n", (u_long)argp->metalsn.file, (u_long)argp->metalsn.offset); printf("\n"); - __db_free(argp); + __os_free(argp, 0); return (0); } @@ -526,11 +524,12 @@ __ham_splitmeta_read(recbuf, argpp) { __ham_splitmeta_args *argp; u_int8_t *bp; + int ret; - argp = (__ham_splitmeta_args *)__db_malloc(sizeof(__ham_splitmeta_args) + - sizeof(DB_TXN)); - if (argp == NULL) - return (ENOMEM); + ret = __os_malloc(sizeof(__ham_splitmeta_args) + + sizeof(DB_TXN), NULL, &argp); + if (ret != 0) + return (ret); argp->txnid = (DB_TXN *)&argp[1]; bp = recbuf; memcpy(&argp->type, bp, sizeof(argp->type)); @@ -581,8 +580,7 @@ int __ham_splitdata_log(logp, txnid, ret_lsnp, flags, rectype = DB_ham_splitdata; txn_num = txnid == NULL ? 0 : txnid->txnid; if (txnid == NULL) { - null_lsn.file = 0; - null_lsn.offset = 0; + ZERO_LSN(null_lsn); lsnp = &null_lsn; } else lsnp = &txnid->last_lsn; @@ -592,8 +590,8 @@ int __ham_splitdata_log(logp, txnid, ret_lsnp, flags, + sizeof(pgno) + sizeof(u_int32_t) + (pageimage == NULL ? 0 : pageimage->size) + sizeof(*pagelsn); - if ((logrec.data = (void *)__db_malloc(logrec.size)) == NULL) - return (ENOMEM); + if ((ret = __os_malloc(logrec.size, NULL, &logrec.data)) != 0) + return (ret); bp = logrec.data; memcpy(bp, &rectype, sizeof(rectype)); @@ -630,7 +628,7 @@ int __ham_splitdata_log(logp, txnid, ret_lsnp, flags, ret = log_put(logp, ret_lsnp, (DBT *)&logrec, flags); if (txnid != NULL) txnid->last_lsn = *ret_lsnp; - __db_free(logrec.data); + __os_free(logrec.data, 0); return (ret); } @@ -681,7 +679,7 @@ __ham_splitdata_print(notused1, dbtp, lsnp, notused2, notused3) printf("\tpagelsn: [%lu][%lu]\n", (u_long)argp->pagelsn.file, (u_long)argp->pagelsn.offset); printf("\n"); - __db_free(argp); + __os_free(argp, 0); return (0); } @@ -695,11 +693,12 @@ __ham_splitdata_read(recbuf, argpp) { __ham_splitdata_args *argp; u_int8_t *bp; + int ret; - argp = (__ham_splitdata_args *)__db_malloc(sizeof(__ham_splitdata_args) + - sizeof(DB_TXN)); - if (argp == NULL) - return (ENOMEM); + ret = __os_malloc(sizeof(__ham_splitdata_args) + + sizeof(DB_TXN), NULL, &argp); + if (ret != 0) + return (ret); argp->txnid = (DB_TXN *)&argp[1]; bp = recbuf; memcpy(&argp->type, bp, sizeof(argp->type)); @@ -756,8 +755,7 @@ int __ham_replace_log(logp, txnid, ret_lsnp, flags, rectype = DB_ham_replace; txn_num = txnid == NULL ? 0 : txnid->txnid; if (txnid == NULL) { - null_lsn.file = 0; - null_lsn.offset = 0; + ZERO_LSN(null_lsn); lsnp = &null_lsn; } else lsnp = &txnid->last_lsn; @@ -770,8 +768,8 @@ int __ham_replace_log(logp, txnid, ret_lsnp, flags, + sizeof(u_int32_t) + (olditem == NULL ? 0 : olditem->size) + sizeof(u_int32_t) + (newitem == NULL ? 0 : newitem->size) + sizeof(makedup); - if ((logrec.data = (void *)__db_malloc(logrec.size)) == NULL) - return (ENOMEM); + if ((ret = __os_malloc(logrec.size, NULL, &logrec.data)) != 0) + return (ret); bp = logrec.data; memcpy(bp, &rectype, sizeof(rectype)); @@ -822,7 +820,7 @@ int __ham_replace_log(logp, txnid, ret_lsnp, flags, ret = log_put(logp, ret_lsnp, (DBT *)&logrec, flags); if (txnid != NULL) txnid->last_lsn = *ret_lsnp; - __db_free(logrec.data); + __os_free(logrec.data, 0); return (ret); } @@ -884,7 +882,7 @@ __ham_replace_print(notused1, dbtp, lsnp, notused2, notused3) printf("\n"); printf("\tmakedup: %lu\n", (u_long)argp->makedup); printf("\n"); - __db_free(argp); + __os_free(argp, 0); return (0); } @@ -898,11 +896,12 @@ __ham_replace_read(recbuf, argpp) { __ham_replace_args *argp; u_int8_t *bp; + int ret; - argp = (__ham_replace_args *)__db_malloc(sizeof(__ham_replace_args) + - sizeof(DB_TXN)); - if (argp == NULL) - return (ENOMEM); + ret = __os_malloc(sizeof(__ham_replace_args) + + sizeof(DB_TXN), NULL, &argp); + if (ret != 0) + return (ret); argp->txnid = (DB_TXN *)&argp[1]; bp = recbuf; memcpy(&argp->type, bp, sizeof(argp->type)); @@ -968,8 +967,7 @@ int __ham_newpgno_log(logp, txnid, ret_lsnp, flags, rectype = DB_ham_newpgno; txn_num = txnid == NULL ? 0 : txnid->txnid; if (txnid == NULL) { - null_lsn.file = 0; - null_lsn.offset = 0; + ZERO_LSN(null_lsn); lsnp = &null_lsn; } else lsnp = &txnid->last_lsn; @@ -983,8 +981,8 @@ int __ham_newpgno_log(logp, txnid, ret_lsnp, flags, + sizeof(new_type) + sizeof(*pagelsn) + sizeof(*metalsn); - if ((logrec.data = (void *)__db_malloc(logrec.size)) == NULL) - return (ENOMEM); + if ((ret = __os_malloc(logrec.size, NULL, &logrec.data)) != 0) + return (ret); bp = logrec.data; memcpy(bp, &rectype, sizeof(rectype)); @@ -1024,7 +1022,7 @@ int __ham_newpgno_log(logp, txnid, ret_lsnp, flags, ret = log_put(logp, ret_lsnp, (DBT *)&logrec, flags); if (txnid != NULL) txnid->last_lsn = *ret_lsnp; - __db_free(logrec.data); + __os_free(logrec.data, 0); return (ret); } @@ -1072,7 +1070,7 @@ __ham_newpgno_print(notused1, dbtp, lsnp, notused2, notused3) printf("\tmetalsn: [%lu][%lu]\n", (u_long)argp->metalsn.file, (u_long)argp->metalsn.offset); printf("\n"); - __db_free(argp); + __os_free(argp, 0); return (0); } @@ -1086,11 +1084,12 @@ __ham_newpgno_read(recbuf, argpp) { __ham_newpgno_args *argp; u_int8_t *bp; + int ret; - argp = (__ham_newpgno_args *)__db_malloc(sizeof(__ham_newpgno_args) + - sizeof(DB_TXN)); - if (argp == NULL) - return (ENOMEM); + ret = __os_malloc(sizeof(__ham_newpgno_args) + + sizeof(DB_TXN), NULL, &argp); + if (ret != 0) + return (ret); argp->txnid = (DB_TXN *)&argp[1]; bp = recbuf; memcpy(&argp->type, bp, sizeof(argp->type)); @@ -1149,8 +1148,7 @@ int __ham_ovfl_log(logp, txnid, ret_lsnp, flags, rectype = DB_ham_ovfl; txn_num = txnid == NULL ? 0 : txnid->txnid; if (txnid == NULL) { - null_lsn.file = 0; - null_lsn.offset = 0; + ZERO_LSN(null_lsn); lsnp = &null_lsn; } else lsnp = &txnid->last_lsn; @@ -1161,8 +1159,8 @@ int __ham_ovfl_log(logp, txnid, ret_lsnp, flags, + sizeof(free_pgno) + sizeof(ovflpoint) + sizeof(*metalsn); - if ((logrec.data = (void *)__db_malloc(logrec.size)) == NULL) - return (ENOMEM); + if ((ret = __os_malloc(logrec.size, NULL, &logrec.data)) != 0) + return (ret); bp = logrec.data; memcpy(bp, &rectype, sizeof(rectype)); @@ -1193,7 +1191,7 @@ int __ham_ovfl_log(logp, txnid, ret_lsnp, flags, ret = log_put(logp, ret_lsnp, (DBT *)&logrec, flags); if (txnid != NULL) txnid->last_lsn = *ret_lsnp; - __db_free(logrec.data); + __os_free(logrec.data, 0); return (ret); } @@ -1237,7 +1235,7 @@ __ham_ovfl_print(notused1, dbtp, lsnp, notused2, notused3) printf("\tmetalsn: [%lu][%lu]\n", (u_long)argp->metalsn.file, (u_long)argp->metalsn.offset); printf("\n"); - __db_free(argp); + __os_free(argp, 0); return (0); } @@ -1251,11 +1249,12 @@ __ham_ovfl_read(recbuf, argpp) { __ham_ovfl_args *argp; u_int8_t *bp; + int ret; - argp = (__ham_ovfl_args *)__db_malloc(sizeof(__ham_ovfl_args) + - sizeof(DB_TXN)); - if (argp == NULL) - return (ENOMEM); + ret = __os_malloc(sizeof(__ham_ovfl_args) + + sizeof(DB_TXN), NULL, &argp); + if (ret != 0) + return (ret); argp->txnid = (DB_TXN *)&argp[1]; bp = recbuf; memcpy(&argp->type, bp, sizeof(argp->type)); @@ -1312,8 +1311,7 @@ int __ham_copypage_log(logp, txnid, ret_lsnp, flags, rectype = DB_ham_copypage; txn_num = txnid == NULL ? 0 : txnid->txnid; if (txnid == NULL) { - null_lsn.file = 0; - null_lsn.offset = 0; + ZERO_LSN(null_lsn); lsnp = &null_lsn; } else lsnp = &txnid->last_lsn; @@ -1326,8 +1324,8 @@ int __ham_copypage_log(logp, txnid, ret_lsnp, flags, + sizeof(nnext_pgno) + sizeof(*nnextlsn) + sizeof(u_int32_t) + (page == NULL ? 0 : page->size); - if ((logrec.data = (void *)__db_malloc(logrec.size)) == NULL) - return (ENOMEM); + if ((ret = __os_malloc(logrec.size, NULL, &logrec.data)) != 0) + return (ret); bp = logrec.data; memcpy(bp, &rectype, sizeof(rectype)); @@ -1376,7 +1374,7 @@ int __ham_copypage_log(logp, txnid, ret_lsnp, flags, ret = log_put(logp, ret_lsnp, (DBT *)&logrec, flags); if (txnid != NULL) txnid->last_lsn = *ret_lsnp; - __db_free(logrec.data); + __os_free(logrec.data, 0); return (ret); } @@ -1432,7 +1430,7 @@ __ham_copypage_print(notused1, dbtp, lsnp, notused2, notused3) } printf("\n"); printf("\n"); - __db_free(argp); + __os_free(argp, 0); return (0); } @@ -1446,11 +1444,12 @@ __ham_copypage_read(recbuf, argpp) { __ham_copypage_args *argp; u_int8_t *bp; + int ret; - argp = (__ham_copypage_args *)__db_malloc(sizeof(__ham_copypage_args) + - sizeof(DB_TXN)); - if (argp == NULL) - return (ENOMEM); + ret = __os_malloc(sizeof(__ham_copypage_args) + + sizeof(DB_TXN), NULL, &argp); + if (ret != 0) + return (ret); argp->txnid = (DB_TXN *)&argp[1]; bp = recbuf; memcpy(&argp->type, bp, sizeof(argp->type)); diff --git a/db2/hash/hash_debug.c b/db2/hash/hash_debug.c deleted file mode 100644 index 232906ae34..0000000000 --- a/db2/hash/hash_debug.c +++ /dev/null @@ -1,92 +0,0 @@ -/*- - * See the file LICENSE for redistribution information. - * - * Copyright (c) 1996, 1997, 1998 - * Sleepycat Software. All rights reserved. - */ -/* - * Copyright (c) 1995 - * The President and Fellows of Harvard University. All rights reserved. - * - * This code is derived from software contributed to Berkeley by - * Jeremy Rassen. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: - * 1. Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * 2. Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution. - * 3. All advertising materials mentioning features or use of this software - * must display the following acknowledgement: - * This product includes software developed by the University of - * California, Berkeley and its contributors. - * 4. Neither the name of the University nor the names of its contributors - * may be used to endorse or promote products derived from this software - * without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND - * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE - * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE - * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE - * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL - * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS - * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) - * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT - * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY - * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF - * SUCH DAMAGE. - */ - -#include "config.h" - -#ifndef lint -static const char sccsid[] = "@(#)hash_debug.c 10.6 (Sleepycat) 5/7/98"; -#endif /* not lint */ - -#ifdef DEBUG -/* - * PACKAGE: hashing - * - * DESCRIPTION: - * Debug routines. - * - * ROUTINES: - * - * External - * __dump_bucket - */ -#ifndef NO_SYSTEM_INCLUDES -#include <sys/types.h> -#endif - -#include "db_int.h" -#include "db_page.h" -#include "hash.h" - -/* - * __ham_dump_bucket -- - * - * PUBLIC: #ifdef DEBUG - * PUBLIC: void __ham_dump_bucket __P((HTAB *, u_int32_t)); - * PUBLIC: #endif - */ -void -__ham_dump_bucket(hashp, bucket) - HTAB *hashp; - u_int32_t bucket; -{ - PAGE *p; - db_pgno_t pgno; - - for (pgno = BUCKET_TO_PAGE(hashp, bucket); pgno != PGNO_INVALID;) { - if (memp_fget(hashp->dbp->mpf, &pgno, 0, &p) != 0) - break; - (void)__db_prpage(p, 1); - pgno = p->next_pgno; - (void)memp_fput(hashp->dbp->mpf, p, 0); - } -} -#endif /* DEBUG */ diff --git a/db2/hash/hash_dup.c b/db2/hash/hash_dup.c index ba248ddb17..bb3466428d 100644 --- a/db2/hash/hash_dup.c +++ b/db2/hash/hash_dup.c @@ -42,7 +42,7 @@ #include "config.h" #ifndef lint -static const char sccsid[] = "@(#)hash_dup.c 10.14 (Sleepycat) 5/7/98"; +static const char sccsid[] = "@(#)hash_dup.c 10.27 (Sleepycat) 12/6/98"; #endif /* not lint */ /* @@ -61,15 +61,17 @@ static const char sccsid[] = "@(#)hash_dup.c 10.14 (Sleepycat) 5/7/98"; #ifndef NO_SYSTEM_INCLUDES #include <sys/types.h> +#include <errno.h> #include <string.h> #endif #include "db_int.h" #include "db_page.h" #include "hash.h" +#include "btree.h" -static int __ham_check_move __P((HTAB *, HASH_CURSOR *, int32_t)); -static int __ham_dup_convert __P((HTAB *, HASH_CURSOR *)); +static int __ham_check_move __P((DBC *, int32_t)); +static int __ham_dup_convert __P((DBC *)); static int __ham_make_dup __P((const DBT *, DBT *d, void **, u_int32_t *)); /* @@ -85,26 +87,29 @@ static int __ham_make_dup __P((const DBT *, DBT *d, void **, u_int32_t *)); * Case 4: The element is large enough to push the duplicate set onto a * separate page. * - * PUBLIC: int __ham_add_dup __P((HTAB *, HASH_CURSOR *, DBT *, u_int32_t)); + * PUBLIC: int __ham_add_dup __P((DBC *, DBT *, u_int32_t)); */ int -__ham_add_dup(hashp, hcp, nval, flags) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_add_dup(dbc, nval, flags) + DBC *dbc; DBT *nval; u_int32_t flags; { - DBT pval, tmp_val; + DB *dbp; + HASH_CURSOR *hcp; + DBT dbt, pval, tmp_val; u_int32_t del_len, new_size; - int ret; + int cmp, ret; u_int8_t *hk; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; if (flags == DB_CURRENT && hcp->dpgno == PGNO_INVALID) del_len = hcp->dup_len; else del_len = 0; - if ((ret = __ham_check_move(hashp, hcp, + if ((ret = __ham_check_move(dbc, (int32_t)DUP_SIZE(nval->size) - (int32_t)del_len)) != 0) return (ret); @@ -117,7 +122,7 @@ __ham_add_dup(hashp, hcp, nval, flags) */ hk = H_PAIRDATA(hcp->pagep, hcp->bndx); new_size = DUP_SIZE(nval->size) - del_len + LEN_HKEYDATA(hcp->pagep, - hashp->hdr->pagesize, H_DATAINDEX(hcp->bndx)); + hcp->hdr->pagesize, H_DATAINDEX(hcp->bndx)); /* * We convert to off-page duplicates if the item is a big item, @@ -125,10 +130,10 @@ __ham_add_dup(hashp, hcp, nval, flags) * if there isn't enough room on this page to add the next item. */ if (HPAGE_PTYPE(hk) != H_OFFDUP && - (HPAGE_PTYPE(hk) == H_OFFPAGE || ISBIG(hashp, new_size) || + (HPAGE_PTYPE(hk) == H_OFFPAGE || ISBIG(hcp, new_size) || DUP_SIZE(nval->size) - del_len > P_FREESPACE(hcp->pagep))) { - if ((ret = __ham_dup_convert(hashp, hcp)) != 0) + if ((ret = __ham_dup_convert(dbc)) != 0) return (ret); else hk = H_PAIRDATA(hcp->pagep, hcp->bndx); @@ -140,30 +145,44 @@ __ham_add_dup(hashp, hcp, nval, flags) HPAGE_PTYPE(hk) = H_DUPLICATE; pval.flags = 0; pval.data = HKEYDATA_DATA(hk); - pval.size = LEN_HDATA(hcp->pagep, hashp->hdr->pagesize, + pval.size = LEN_HDATA(hcp->pagep, dbp->pgsize, hcp->bndx); if ((ret = - __ham_make_dup(&pval, &tmp_val, &hcp->big_data, - &hcp->big_datalen)) != 0 || (ret = - __ham_replpair(hashp, hcp, &tmp_val, 1)) != 0) + __ham_make_dup(&pval, &tmp_val, &dbc->rdata.data, + &dbc->rdata.size)) != 0 || (ret = + __ham_replpair(dbc, &tmp_val, 1)) != 0) return (ret); } /* Now make the new entry a duplicate. */ if ((ret = __ham_make_dup(nval, - &tmp_val, &hcp->big_data, &hcp->big_datalen)) != 0) + &tmp_val, &dbc->rdata.data, &dbc->rdata.size)) != 0) return (ret); tmp_val.dlen = 0; switch (flags) { /* On page. */ case DB_KEYFIRST: - tmp_val.doff = 0; - break; case DB_KEYLAST: - tmp_val.doff = LEN_HDATA(hcp->pagep, - hashp->hdr->pagesize, hcp->bndx); + if (dbp->dup_compare != NULL) + __ham_dsearch(dbc, nval, &tmp_val.doff, &cmp); + else if (flags == DB_KEYFIRST) + tmp_val.doff = 0; + else + tmp_val.doff = LEN_HDATA(hcp->pagep, + hcp->hdr->pagesize, hcp->bndx); break; case DB_CURRENT: + /* + * If we have a sort function, we need to verify that + * the new item sorts identically to the old item. + */ + if (dbp->dup_compare != NULL) { + dbt.data = HKEYDATA_DATA(H_PAIRDATA(hcp->pagep, + hcp->bndx)) + hcp->dup_off; + dbt.size = DUP_SIZE(hcp->dup_len); + if (dbp->dup_compare(nval, &dbt) != 0) + return (EINVAL); + } tmp_val.doff = hcp->dup_off; tmp_val.dlen = DUP_SIZE(hcp->dup_len); break; @@ -175,9 +194,9 @@ __ham_add_dup(hashp, hcp, nval, flags) break; } /* Add the duplicate. */ - ret = __ham_replpair(hashp, hcp, &tmp_val, 0); + ret = __ham_replpair(dbc, &tmp_val, 0); if (ret == 0) - ret = __ham_dirty_page(hashp, hcp->pagep); + ret = __ham_dirty_page(dbp, hcp->pagep); __ham_c_update(hcp, hcp->pgno, tmp_val.size, 1, 1); return (ret); } @@ -190,27 +209,48 @@ __ham_add_dup(hashp, hcp, nval, flags) switch (flags) { case DB_KEYFIRST: + if (dbp->dup_compare != NULL) + goto sorted_dups; /* * The only way that we are already on a dup page is * if we just converted the on-page representation. * In that case, we've only got one page of duplicates. */ if (hcp->dpagep == NULL && (ret = - __db_dend(hashp->dbp, hcp->dpgno, &hcp->dpagep)) != 0) + __db_dend(dbc, hcp->dpgno, &hcp->dpagep)) != 0) return (ret); hcp->dndx = 0; break; case DB_KEYLAST: - if (hcp->dpagep == NULL && (ret = - __db_dend(hashp->dbp, hcp->dpgno, &hcp->dpagep)) != 0) - return (ret); - hcp->dpgno = PGNO(hcp->dpagep); - hcp->dndx = NUM_ENT(hcp->dpagep); + if (dbp->dup_compare != NULL) { +sorted_dups: if ((ret = __db_dsearch(dbc, 1, nval, + hcp->dpgno, &hcp->dndx, &hcp->dpagep, &cmp)) != 0) + return (ret); + if (cmp == 0) + hcp->dpgno = PGNO(hcp->dpagep); + } else { + if (hcp->dpagep == NULL && (ret = + __db_dend(dbc, hcp->dpgno, &hcp->dpagep)) != 0) + return (ret); + hcp->dpgno = PGNO(hcp->dpagep); + hcp->dndx = NUM_ENT(hcp->dpagep); + } break; case DB_CURRENT: - if ((ret = __db_ditem(hashp->dbp, hcp->dpagep, hcp->dndx, - BKEYDATA_SIZE(GET_BKEYDATA(hcp->dpagep, hcp->dndx)->len))) - != 0) + if (dbp->dup_compare != NULL && __bam_cmp(dbp, + nval, hcp->dpagep, hcp->dndx, dbp->dup_compare) != 0) + return (EINVAL); + switch (GET_BKEYDATA(hcp->dpagep, hcp->dndx)->type) { + case B_KEYDATA: + del_len = BKEYDATA_SIZE(GET_BKEYDATA(hcp->dpagep, + hcp->dndx)->len); + break; + case B_OVERFLOW: + del_len = BOVERFLOW_SIZE; + break; + } + if ((ret = + __db_ditem(dbc, hcp->dpagep, hcp->dndx, del_len)) != 0) return (ret); break; case DB_BEFORE: /* The default behavior is correct. */ @@ -220,7 +260,7 @@ __ham_add_dup(hashp, hcp, nval, flags) break; } - ret = __db_dput(hashp->dbp, + ret = __db_dput(dbc, nval, &hcp->dpagep, &hcp->dndx, __ham_overflow_page); hcp->pgno = PGNO(hcp->pagep); __ham_c_update(hcp, hcp->pgno, nval->size, 1, 1); @@ -231,22 +271,25 @@ __ham_add_dup(hashp, hcp, nval, flags) * Convert an on-page set of duplicates to an offpage set of duplicates. */ static int -__ham_dup_convert(hashp, hcp) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_dup_convert(dbc) + DBC *dbc; { + DB *dbp; + HASH_CURSOR *hcp; BOVERFLOW bo; DBT dbt; HOFFPAGE ho; - db_indx_t dndx, len; + db_indx_t dndx, i, len, off; int ret; u_int8_t *p, *pend; /* * Create a new page for the duplicates. */ + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; if ((ret = - __ham_overflow_page(hashp->dbp, P_DUPLICATE, &hcp->dpagep)) != 0) + __ham_overflow_page(dbc, P_DUPLICATE, &hcp->dpagep)) != 0) return (ret); hcp->dpagep->type = P_DUPLICATE; hcp->dpgno = PGNO(hcp->dpagep); @@ -254,67 +297,80 @@ __ham_dup_convert(hashp, hcp) /* * Now put the duplicates onto the new page. */ + dndx = 0; dbt.flags = 0; switch (HPAGE_PTYPE(H_PAIRDATA(hcp->pagep, hcp->bndx))) { case H_KEYDATA: /* Simple case, one key on page; move it to dup page. */ - dndx = 0; dbt.size = - LEN_HDATA(hcp->pagep, hashp->hdr->pagesize, hcp->bndx); + LEN_HDATA(hcp->pagep, hcp->hdr->pagesize, hcp->bndx); dbt.data = HKEYDATA_DATA(H_PAIRDATA(hcp->pagep, hcp->bndx)); - ret = __db_pitem(hashp->dbp, hcp->dpagep, + ret = __db_pitem(dbc, hcp->dpagep, (u_int32_t)dndx, BKEYDATA_SIZE(dbt.size), NULL, &dbt); if (ret == 0) - __ham_dirty_page(hashp, hcp->dpagep); + __ham_dirty_page(dbp, hcp->dpagep); break; case H_OFFPAGE: /* Simple case, one key on page; move it to dup page. */ - dndx = 0; memcpy(&ho, P_ENTRY(hcp->pagep, H_DATAINDEX(hcp->bndx)), HOFFPAGE_SIZE); + UMRW(bo.unused1); B_TSET(bo.type, ho.type, 0); + UMRW(bo.unused2); bo.pgno = ho.pgno; bo.tlen = ho.tlen; dbt.size = BOVERFLOW_SIZE; dbt.data = &bo; - ret = __db_pitem(hashp->dbp, hcp->dpagep, + ret = __db_pitem(dbc, hcp->dpagep, (u_int32_t)dndx, dbt.size, &dbt, NULL); if (ret == 0) - __ham_dirty_page(hashp, hcp->dpagep); + __ham_dirty_page(dbp, hcp->dpagep); break; case H_DUPLICATE: p = HKEYDATA_DATA(H_PAIRDATA(hcp->pagep, hcp->bndx)); pend = p + - LEN_HDATA(hcp->pagep, hashp->hdr->pagesize, hcp->bndx); + LEN_HDATA(hcp->pagep, hcp->hdr->pagesize, hcp->bndx); - for (dndx = 0; p < pend; dndx++) { + /* + * We need to maintain the duplicate cursor position. + * Keep track of where we are in the duplicate set via + * the offset, and when it matches the one in the cursor, + * set the off-page duplicate cursor index to the current + * index. + */ + for (off = 0, i = 0; p < pend; i++) { + if (off == hcp->dup_off) + dndx = i; memcpy(&len, p, sizeof(db_indx_t)); dbt.size = len; p += sizeof(db_indx_t); dbt.data = p; p += len + sizeof(db_indx_t); - ret = __db_dput(hashp->dbp, &dbt, - &hcp->dpagep, &dndx, __ham_overflow_page); + off += len + 2 * sizeof(db_indx_t); + ret = __db_dput(dbc, &dbt, + &hcp->dpagep, &i, __ham_overflow_page); if (ret != 0) break; } break; default: - ret = __db_pgfmt(hashp->dbp, (u_long)hcp->pgno); + ret = __db_pgfmt(dbp, (u_long)hcp->pgno); + break; } if (ret == 0) { /* * Now attach this to the source page in place of * the old duplicate item. */ - __ham_move_offpage(hashp, hcp->pagep, + __ham_move_offpage(dbc, hcp->pagep, (u_int32_t)H_DATAINDEX(hcp->bndx), hcp->dpgno); /* Can probably just do a "put" here. */ - ret = __ham_dirty_page(hashp, hcp->pagep); + ret = __ham_dirty_page(dbp, hcp->pagep); + hcp->dndx = dndx; } else { - (void)__ham_del_page(hashp->dbp, hcp->dpagep); + (void)__ham_del_page(dbc, hcp->dpagep); hcp->dpagep = NULL; } return (ret); @@ -354,11 +410,12 @@ __ham_make_dup(notdup, duplicate, bufp, sizep) } static int -__ham_check_move(hashp, hcp, add_len) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_check_move(dbc, add_len) + DBC *dbc; int32_t add_len; { + DB *dbp; + HASH_CURSOR *hcp; DBT k, d; DB_LSN new_lsn; PAGE *next_pagep; @@ -367,6 +424,8 @@ __ham_check_move(hashp, hcp, add_len) u_int8_t *hk; int ret; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; /* * Check if we can do whatever we need to on this page. If not, * then we'll have to move the current element to a new page. @@ -381,7 +440,7 @@ __ham_check_move(hashp, hcp, add_len) return (0); old_len = - LEN_HITEM(hcp->pagep, hashp->hdr->pagesize, H_DATAINDEX(hcp->bndx)); + LEN_HITEM(hcp->pagep, hcp->hdr->pagesize, H_DATAINDEX(hcp->bndx)); new_datalen = old_len - HKEYDATA_SIZE(0) + add_len; /* @@ -392,11 +451,11 @@ __ham_check_move(hashp, hcp, add_len) * threshold, but the new data won't fit on the page. * If neither of these is true, then we can return. */ - if (ISBIG(hashp, new_datalen) && (old_len > HOFFDUP_SIZE || + if (ISBIG(hcp, new_datalen) && (old_len > HOFFDUP_SIZE || HOFFDUP_SIZE - old_len <= P_FREESPACE(hcp->pagep))) return (0); - if (!ISBIG(hashp, new_datalen) && + if (!ISBIG(hcp, new_datalen) && add_len <= (int32_t)P_FREESPACE(hcp->pagep)) return (0); @@ -405,18 +464,18 @@ __ham_check_move(hashp, hcp, add_len) * Check if there are more pages in the chain. */ - new_datalen = ISBIG(hashp, new_datalen) ? + new_datalen = ISBIG(hcp, new_datalen) ? HOFFDUP_SIZE : HKEYDATA_SIZE(new_datalen); next_pagep = NULL; for (next_pgno = NEXT_PGNO(hcp->pagep); next_pgno != PGNO_INVALID; next_pgno = NEXT_PGNO(next_pagep)) { if (next_pagep != NULL && - (ret = __ham_put_page(hashp->dbp, next_pagep, 0)) != 0) + (ret = __ham_put_page(dbp, next_pagep, 0)) != 0) return (ret); if ((ret = - __ham_get_page(hashp->dbp, next_pgno, &next_pagep)) != 0) + __ham_get_page(dbp, next_pgno, &next_pagep)) != 0) return (ret); if (P_FREESPACE(next_pagep) >= new_datalen) @@ -424,17 +483,17 @@ __ham_check_move(hashp, hcp, add_len) } /* No more pages, add one. */ - if (next_pagep == NULL && - (ret = __ham_add_ovflpage(hashp, hcp->pagep, 0, &next_pagep)) != 0) + if (next_pagep == NULL && (ret = __ham_add_ovflpage(dbc, + hcp->pagep, 0, &next_pagep)) != 0) return (ret); /* Add new page at the end of the chain. */ - if (P_FREESPACE(next_pagep) < new_datalen && - (ret = __ham_add_ovflpage(hashp, next_pagep, 1, &next_pagep)) != 0) + if (P_FREESPACE(next_pagep) < new_datalen && (ret = + __ham_add_ovflpage(dbc, next_pagep, 1, &next_pagep)) != 0) return (ret); /* Copy the item to the new page. */ - if (DB_LOGGING(hashp->dbp)) { + if (DB_LOGGING(hcp->dbc)) { rectype = PUTPAIR; k.flags = 0; d.flags = 0; @@ -447,7 +506,7 @@ __ham_check_move(hashp, hcp, add_len) k.data = HKEYDATA_DATA(H_PAIRKEY(hcp->pagep, hcp->bndx)); k.size = LEN_HKEY(hcp->pagep, - hashp->hdr->pagesize, hcp->bndx); + hcp->hdr->pagesize, hcp->bndx); } if (HPAGE_PTYPE(hk) == H_OFFPAGE) { @@ -458,13 +517,13 @@ __ham_check_move(hashp, hcp, add_len) d.data = HKEYDATA_DATA(H_PAIRDATA(hcp->pagep, hcp->bndx)); d.size = LEN_HDATA(hcp->pagep, - hashp->hdr->pagesize, hcp->bndx); + hcp->hdr->pagesize, hcp->bndx); } - if ((ret = __ham_insdel_log(hashp->dbp->dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, rectype, - hashp->dbp->log_fileid, PGNO(next_pagep), + if ((ret = __ham_insdel_log(dbp->dbenv->lg_info, + dbc->txn, &new_lsn, 0, rectype, + dbp->log_fileid, PGNO(next_pagep), (u_int32_t)H_NUMPAIRS(next_pagep), &LSN(next_pagep), &k, &d)) != 0) return (ret); @@ -473,13 +532,15 @@ __ham_check_move(hashp, hcp, add_len) LSN(next_pagep) = new_lsn; /* Structure assignment. */ } - __ham_copy_item(hashp, hcp->pagep, H_KEYINDEX(hcp->bndx), next_pagep); - __ham_copy_item(hashp, hcp->pagep, H_DATAINDEX(hcp->bndx), next_pagep); + __ham_copy_item(dbp->pgsize, + hcp->pagep, H_KEYINDEX(hcp->bndx), next_pagep); + __ham_copy_item(dbp->pgsize, + hcp->pagep, H_DATAINDEX(hcp->bndx), next_pagep); /* Now delete the pair from the current page. */ - ret = __ham_del_pair(hashp, hcp, 0); + ret = __ham_del_pair(dbc, 0); - (void)__ham_put_page(hashp->dbp, hcp->pagep, 1); + (void)__ham_put_page(dbp, hcp->pagep, 1); hcp->pagep = next_pagep; hcp->pgno = PGNO(hcp->pagep); hcp->bndx = H_NUMPAIRS(hcp->pagep) - 1; @@ -488,19 +549,25 @@ __ham_check_move(hashp, hcp, add_len) } /* - * Replace an onpage set of duplicates with the OFFDUP structure that - * references the duplicate page. - * XXX This is really just a special case of __onpage_replace; we should + * __ham_move_offpage -- + * Replace an onpage set of duplicates with the OFFDUP structure + * that references the duplicate page. + * + * XXX + * This is really just a special case of __onpage_replace; we should * probably combine them. - * PUBLIC: void __ham_move_offpage __P((HTAB *, PAGE *, u_int32_t, db_pgno_t)); + * + * PUBLIC: void __ham_move_offpage __P((DBC *, PAGE *, u_int32_t, db_pgno_t)); */ void -__ham_move_offpage(hashp, pagep, ndx, pgno) - HTAB *hashp; +__ham_move_offpage(dbc, pagep, ndx, pgno) + DBC *dbc; PAGE *pagep; u_int32_t ndx; db_pgno_t pgno; { + DB *dbp; + HASH_CURSOR *hcp; DBT new_dbt; DBT old_dbt; HOFFDUP od; @@ -508,22 +575,27 @@ __ham_move_offpage(hashp, pagep, ndx, pgno) int32_t shrink; u_int8_t *src; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; od.type = H_OFFDUP; + UMRW(od.unused[0]); + UMRW(od.unused[1]); + UMRW(od.unused[2]); od.pgno = pgno; - if (DB_LOGGING(hashp->dbp)) { + if (DB_LOGGING(dbc)) { new_dbt.data = &od; new_dbt.size = HOFFDUP_SIZE; old_dbt.data = P_ENTRY(pagep, ndx); - old_dbt.size = LEN_HITEM(pagep, hashp->hdr->pagesize, ndx); - (void)__ham_replace_log(hashp->dbp->dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &LSN(pagep), 0, - hashp->dbp->log_fileid, PGNO(pagep), (u_int32_t)ndx, - &LSN(pagep), -1, &old_dbt, &new_dbt, 0); + old_dbt.size = LEN_HITEM(pagep, hcp->hdr->pagesize, ndx); + (void)__ham_replace_log(dbp->dbenv->lg_info, + dbc->txn, &LSN(pagep), 0, dbp->log_fileid, + PGNO(pagep), (u_int32_t)ndx, &LSN(pagep), -1, + &old_dbt, &new_dbt, 0); } shrink = - LEN_HITEM(pagep, hashp->hdr->pagesize, ndx) - HOFFDUP_SIZE; + LEN_HITEM(pagep, hcp->hdr->pagesize, ndx) - HOFFDUP_SIZE; if (shrink != 0) { /* Copy data. */ @@ -539,3 +611,46 @@ __ham_move_offpage(hashp, pagep, ndx, pgno) /* Now copy the offdup entry onto the page. */ memcpy(P_ENTRY(pagep, ndx), &od, HOFFDUP_SIZE); } + +/* + * __ham_dsearch: + * Locate a particular duplicate in a duplicate set. + * + * PUBLIC: void __ham_dsearch __P((DBC *, DBT *, u_int32_t *, int *)); + */ +void +__ham_dsearch(dbc, dbt, offp, cmpp) + DBC *dbc; + DBT *dbt; + u_int32_t *offp; + int *cmpp; +{ + DB *dbp; + HASH_CURSOR *hcp; + DBT cur; + db_indx_t i, len; + int (*func) __P((const DBT *, const DBT *)); + u_int8_t *data; + + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; + if (dbp->dup_compare == NULL) + func = __bam_defcmp; + else + func = dbp->dup_compare; + + i = F_ISSET(dbc, DBC_CONTINUE) ? hcp->dup_off: 0; + data = HKEYDATA_DATA(H_PAIRDATA(hcp->pagep, hcp->bndx)) + i; + while (i < LEN_HDATA(hcp->pagep, hcp->hdr->pagesize, hcp->bndx)) { + memcpy(&len, data, sizeof(db_indx_t)); + data += sizeof(db_indx_t); + cur.data = data; + cur.size = (u_int32_t)len; + *cmpp = func(dbt, &cur); + if (*cmpp == 0 || (*cmpp < 0 && dbp->dup_compare != NULL)) + break; + i += len + 2 * sizeof(db_indx_t); + data += len + sizeof(db_indx_t); + } + *offp = i; +} diff --git a/db2/hash/hash_page.c b/db2/hash/hash_page.c index 5b3463947b..3419c1215c 100644 --- a/db2/hash/hash_page.c +++ b/db2/hash/hash_page.c @@ -47,7 +47,7 @@ #include "config.h" #ifndef lint -static const char sccsid[] = "@(#)hash_page.c 10.40 (Sleepycat) 6/2/98"; +static const char sccsid[] = "@(#)hash_page.c 10.55 (Sleepycat) 1/3/99"; #endif /* not lint */ /* @@ -77,107 +77,118 @@ static const char sccsid[] = "@(#)hash_page.c 10.40 (Sleepycat) 6/2/98"; #include "db_page.h" #include "hash.h" -static int __ham_lock_bucket __P((DB *, HASH_CURSOR *, db_lockmode_t)); +static int __ham_lock_bucket __P((DBC *, db_lockmode_t)); #ifdef DEBUG_SLOW -static void __account_page(HTAB *, db_pgno_t, int); +static void __account_page(DB *, db_pgno_t, int); #endif /* - * PUBLIC: int __ham_item __P((HTAB *, HASH_CURSOR *, db_lockmode_t)); + * PUBLIC: int __ham_item __P((DBC *, db_lockmode_t)); */ int -__ham_item(hashp, cursorp, mode) - HTAB *hashp; - HASH_CURSOR *cursorp; +__ham_item(dbc, mode) + DBC *dbc; db_lockmode_t mode; { + DB *dbp; + HASH_CURSOR *hcp; db_pgno_t next_pgno; int ret; - if (F_ISSET(cursorp, H_DELETED)) + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; + + if (F_ISSET(hcp, H_DELETED)) return (EINVAL); - F_CLR(cursorp, H_OK | H_NOMORE); + F_CLR(hcp, H_OK | H_NOMORE); /* Check if we need to get a page for this cursor. */ - if ((ret = __ham_get_cpage(hashp, cursorp, mode)) != 0) + if ((ret = __ham_get_cpage(dbc, mode)) != 0) return (ret); /* Check if we are looking for space in which to insert an item. */ - if (cursorp->seek_size && cursorp->seek_found_page == PGNO_INVALID - && cursorp->seek_size < P_FREESPACE(cursorp->pagep)) - cursorp->seek_found_page = cursorp->pgno; + if (hcp->seek_size && hcp->seek_found_page == PGNO_INVALID + && hcp->seek_size < P_FREESPACE(hcp->pagep)) + hcp->seek_found_page = hcp->pgno; /* Check if we need to go on to the next page. */ - if (F_ISSET(cursorp, H_ISDUP) && cursorp->dpgno == PGNO_INVALID) + if (F_ISSET(hcp, H_ISDUP) && hcp->dpgno == PGNO_INVALID) /* * ISDUP is set, and offset is at the beginning of the datum. * We need to grab the length of the datum, then set the datum * pointer to be the beginning of the datum. */ - memcpy(&cursorp->dup_len, - HKEYDATA_DATA(H_PAIRDATA(cursorp->pagep, cursorp->bndx)) + - cursorp->dup_off, sizeof(db_indx_t)); - else if (F_ISSET(cursorp, H_ISDUP)) { + memcpy(&hcp->dup_len, + HKEYDATA_DATA(H_PAIRDATA(hcp->pagep, hcp->bndx)) + + hcp->dup_off, sizeof(db_indx_t)); + else if (F_ISSET(hcp, H_ISDUP)) { /* Make sure we're not about to run off the page. */ - if (cursorp->dpagep == NULL && (ret = __ham_get_page(hashp->dbp, - cursorp->dpgno, &cursorp->dpagep)) != 0) + if (hcp->dpagep == NULL && (ret = __ham_get_page(dbp, + hcp->dpgno, &hcp->dpagep)) != 0) return (ret); - if (cursorp->dndx >= NUM_ENT(cursorp->dpagep)) { - if (NEXT_PGNO(cursorp->dpagep) == PGNO_INVALID) { - if ((ret = __ham_put_page(hashp->dbp, - cursorp->dpagep, 0)) != 0) + if (hcp->dndx >= NUM_ENT(hcp->dpagep)) { + if (NEXT_PGNO(hcp->dpagep) == PGNO_INVALID) { + if (F_ISSET(hcp, H_DUPONLY)) { + F_CLR(hcp, H_OK); + F_SET(hcp, H_NOMORE); + return (0); + } + if ((ret = __ham_put_page(dbp, + hcp->dpagep, 0)) != 0) return (ret); - F_CLR(cursorp, H_ISDUP); - cursorp->dpagep = NULL; - cursorp->dpgno = PGNO_INVALID; - cursorp->dndx = NDX_INVALID; - cursorp->bndx++; - } else if ((ret = __ham_next_cpage(hashp, cursorp, - NEXT_PGNO(cursorp->dpagep), 0, H_ISDUP)) != 0) + F_CLR(hcp, H_ISDUP); + hcp->dpagep = NULL; + hcp->dpgno = PGNO_INVALID; + hcp->dndx = NDX_INVALID; + hcp->bndx++; + } else if ((ret = __ham_next_cpage(dbc, + NEXT_PGNO(hcp->dpagep), 0, H_ISDUP)) != 0) return (ret); } } - if (cursorp->bndx >= (db_indx_t)H_NUMPAIRS(cursorp->pagep)) { + if (hcp->bndx >= (db_indx_t)H_NUMPAIRS(hcp->pagep)) { /* Fetch next page. */ - if (NEXT_PGNO(cursorp->pagep) == PGNO_INVALID) { - F_SET(cursorp, H_NOMORE); - if (cursorp->dpagep != NULL && - (ret = __ham_put_page(hashp->dbp, - cursorp->dpagep, 0)) != 0) + if (NEXT_PGNO(hcp->pagep) == PGNO_INVALID) { + F_SET(hcp, H_NOMORE); + if (hcp->dpagep != NULL && + (ret = __ham_put_page(dbp, hcp->dpagep, 0)) != 0) return (ret); - cursorp->dpgno = PGNO_INVALID; + hcp->dpgno = PGNO_INVALID; return (DB_NOTFOUND); } - next_pgno = NEXT_PGNO(cursorp->pagep); - cursorp->bndx = 0; - if ((ret = __ham_next_cpage(hashp, - cursorp, next_pgno, 0, 0)) != 0) + next_pgno = NEXT_PGNO(hcp->pagep); + hcp->bndx = 0; + if ((ret = __ham_next_cpage(dbc, next_pgno, 0, 0)) != 0) return (ret); } - F_SET(cursorp, H_OK); + F_SET(hcp, H_OK); return (0); } /* - * PUBLIC: int __ham_item_reset __P((HTAB *, HASH_CURSOR *)); + * PUBLIC: int __ham_item_reset __P((DBC *)); */ int -__ham_item_reset(hashp, cursorp) - HTAB *hashp; - HASH_CURSOR *cursorp; +__ham_item_reset(dbc) + DBC *dbc; { + HASH_CURSOR *hcp; + DB *dbp; int ret; - if (cursorp->pagep) - ret = __ham_put_page(hashp->dbp, cursorp->pagep, 0); - else - ret = 0; - - __ham_item_init(cursorp); + ret = 0; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; + if (hcp->pagep != NULL) + ret = __ham_put_page(dbp, hcp->pagep, 0); + if (ret == 0 && hcp->dpagep != NULL) + ret = __ham_put_page(dbp, hcp->dpagep, 0); + + __ham_item_init(hcp); return (ret); } @@ -185,57 +196,67 @@ __ham_item_reset(hashp, cursorp) * PUBLIC: void __ham_item_init __P((HASH_CURSOR *)); */ void -__ham_item_init(cursorp) - HASH_CURSOR *cursorp; +__ham_item_init(hcp) + HASH_CURSOR *hcp; { - cursorp->pagep = NULL; - cursorp->bucket = BUCKET_INVALID; - cursorp->lock = 0; - cursorp->bndx = NDX_INVALID; - cursorp->pgno = PGNO_INVALID; - cursorp->dpgno = PGNO_INVALID; - cursorp->dndx = NDX_INVALID; - cursorp->dpagep = NULL; - cursorp->flags = 0; - cursorp->seek_size = 0; - cursorp->seek_found_page = PGNO_INVALID; + /* + * If this cursor still holds any locks, we must + * release them if we are not running with transactions. + */ + if (hcp->lock && hcp->dbc->txn == NULL) + (void)lock_put(hcp->dbc->dbp->dbenv->lk_info, hcp->lock); + + /* + * The following fields must *not* be initialized here + * because they may have meaning across inits. + * hlock, hdr, split_buf, stats + */ + hcp->bucket = BUCKET_INVALID; + hcp->lbucket = BUCKET_INVALID; + hcp->lock = 0; + hcp->pagep = NULL; + hcp->pgno = PGNO_INVALID; + hcp->bndx = NDX_INVALID; + hcp->dpagep = NULL; + hcp->dpgno = PGNO_INVALID; + hcp->dndx = NDX_INVALID; + hcp->dup_off = 0; + hcp->dup_len = 0; + hcp->dup_tlen = 0; + hcp->seek_size = 0; + hcp->seek_found_page = PGNO_INVALID; + hcp->flags = 0; } /* - * PUBLIC: int __ham_item_done __P((HTAB *, HASH_CURSOR *, int)); + * PUBLIC: int __ham_item_done __P((DBC *, int)); */ int -__ham_item_done(hashp, cursorp, dirty) - HTAB *hashp; - HASH_CURSOR *cursorp; +__ham_item_done(dbc, dirty) + DBC *dbc; int dirty; { + DB *dbp; + HASH_CURSOR *hcp; int ret, t_ret; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; t_ret = ret = 0; - if (cursorp->pagep) - ret = __ham_put_page(hashp->dbp, cursorp->pagep, - dirty && cursorp->dpagep == NULL); - cursorp->pagep = NULL; + if (hcp->pagep) + ret = __ham_put_page(dbp, hcp->pagep, + dirty && hcp->dpagep == NULL); + hcp->pagep = NULL; - if (cursorp->dpagep) - t_ret = __ham_put_page(hashp->dbp, cursorp->dpagep, dirty); - cursorp->dpagep = NULL; + if (hcp->dpagep) + t_ret = __ham_put_page(dbp, hcp->dpagep, dirty); + hcp->dpagep = NULL; if (ret == 0 && t_ret != 0) ret = t_ret; /* - * If we are running with transactions, then we must - * not relinquish locks explicitly. - */ - if (cursorp->lock && hashp->dbp->txn == NULL) - t_ret = lock_put(hashp->dbp->dbenv->lk_info, cursorp->lock); - cursorp->lock = 0; - - - /* * We don't throw out the page number since we might want to * continue getting on this page. */ @@ -245,40 +266,42 @@ __ham_item_done(hashp, cursorp, dirty) /* * Returns the last item in a bucket. * - * PUBLIC: int __ham_item_last __P((HTAB *, HASH_CURSOR *, db_lockmode_t)); + * PUBLIC: int __ham_item_last __P((DBC *, db_lockmode_t)); */ int -__ham_item_last(hashp, cursorp, mode) - HTAB *hashp; - HASH_CURSOR *cursorp; +__ham_item_last(dbc, mode) + DBC *dbc; db_lockmode_t mode; { + HASH_CURSOR *hcp; int ret; - if ((ret = __ham_item_reset(hashp, cursorp)) != 0) + hcp = (HASH_CURSOR *)dbc->internal; + if ((ret = __ham_item_reset(dbc)) != 0) return (ret); - cursorp->bucket = hashp->hdr->max_bucket; - F_SET(cursorp, H_OK); - return (__ham_item_prev(hashp, cursorp, mode)); + hcp->bucket = hcp->hdr->max_bucket; + F_SET(hcp, H_OK); + return (__ham_item_prev(dbc, mode)); } /* - * PUBLIC: int __ham_item_first __P((HTAB *, HASH_CURSOR *, db_lockmode_t)); + * PUBLIC: int __ham_item_first __P((DBC *, db_lockmode_t)); */ int -__ham_item_first(hashp, cursorp, mode) - HTAB *hashp; - HASH_CURSOR *cursorp; +__ham_item_first(dbc, mode) + DBC *dbc; db_lockmode_t mode; { + HASH_CURSOR *hcp; int ret; - if ((ret = __ham_item_reset(hashp, cursorp)) != 0) + hcp = (HASH_CURSOR *)dbc->internal; + if ((ret = __ham_item_reset(dbc)) != 0) return (ret); - F_SET(cursorp, H_OK); - cursorp->bucket = 0; - return (__ham_item_next(hashp, cursorp, mode)); + F_SET(hcp, H_OK); + hcp->bucket = 0; + return (__ham_item_next(dbc, mode)); } /* @@ -287,17 +310,20 @@ __ham_item_first(hashp, cursorp, mode) * bigkeys, just returns the page number and index of the bigkey * pointer pair. * - * PUBLIC: int __ham_item_prev __P((HTAB *, HASH_CURSOR *, db_lockmode_t)); + * PUBLIC: int __ham_item_prev __P((DBC *, db_lockmode_t)); */ int -__ham_item_prev(hashp, cursorp, mode) - HTAB *hashp; - HASH_CURSOR *cursorp; +__ham_item_prev(dbc, mode) + DBC *dbc; db_lockmode_t mode; { + DB *dbp; + HASH_CURSOR *hcp; db_pgno_t next_pgno; int ret; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; /* * There are N cases for backing up in a hash file. * Case 1: In the middle of a page, no duplicates, just dec the index. @@ -307,52 +333,56 @@ __ham_item_prev(hashp, cursorp, mode) * Case 4: At the beginning of a page; go to previous page. * Case 5: At the beginning of a bucket; go to prev bucket. */ - F_CLR(cursorp, H_OK | H_NOMORE | H_DELETED); + F_CLR(hcp, H_OK | H_NOMORE | H_DELETED); /* * First handle the duplicates. Either you'll get the key here * or you'll exit the duplicate set and drop into the code below * to handle backing up through keys. */ - if (F_ISSET(cursorp, H_ISDUP)) { - if (cursorp->dpgno == PGNO_INVALID) { + if (F_ISSET(hcp, H_ISDUP)) { + if (hcp->dpgno == PGNO_INVALID) { /* Duplicates are on-page. */ - if (cursorp->dup_off != 0) { - if ((ret = __ham_get_cpage(hashp, - cursorp, mode)) != 0) + if (hcp->dup_off != 0) { + if ((ret = __ham_get_cpage(dbc, mode)) != 0) return (ret); else { HASH_CURSOR *h; - h = cursorp; + h = hcp; memcpy(&h->dup_len, HKEYDATA_DATA( H_PAIRDATA(h->pagep, h->bndx)) + h->dup_off - sizeof(db_indx_t), sizeof(db_indx_t)); - cursorp->dup_off -= - DUP_SIZE(cursorp->dup_len); - cursorp->dndx--; - return (__ham_item(hashp, - cursorp, mode)); + hcp->dup_off -= + DUP_SIZE(hcp->dup_len); + hcp->dndx--; + return (__ham_item(dbc, mode)); } } - } else if (cursorp->dndx > 0) { /* Duplicates are off-page. */ - cursorp->dndx--; - return (__ham_item(hashp, cursorp, mode)); - } else if ((ret = __ham_get_cpage(hashp, cursorp, mode)) != 0) + } else if (hcp->dndx > 0) { /* Duplicates are off-page. */ + hcp->dndx--; + return (__ham_item(dbc, mode)); + } else if ((ret = __ham_get_cpage(dbc, mode)) != 0) return (ret); - else if (PREV_PGNO(cursorp->dpagep) == PGNO_INVALID) { - F_CLR(cursorp, H_ISDUP); /* End of dups */ - cursorp->dpgno = PGNO_INVALID; - if (cursorp->dpagep != NULL) - (void)__ham_put_page(hashp->dbp, - cursorp->dpagep, 0); - cursorp->dpagep = NULL; - } else if ((ret = __ham_next_cpage(hashp, cursorp, - PREV_PGNO(cursorp->dpagep), 0, H_ISDUP)) != 0) + else if (PREV_PGNO(hcp->dpagep) == PGNO_INVALID) { + if (F_ISSET(hcp, H_DUPONLY)) { + F_CLR(hcp, H_OK); + F_SET(hcp, H_NOMORE); + return (0); + } else { + F_CLR(hcp, H_ISDUP); /* End of dups */ + hcp->dpgno = PGNO_INVALID; + if (hcp->dpagep != NULL) + (void)__ham_put_page(dbp, + hcp->dpagep, 0); + hcp->dpagep = NULL; + } + } else if ((ret = __ham_next_cpage(dbc, + PREV_PGNO(hcp->dpagep), 0, H_ISDUP)) != 0) return (ret); else { - cursorp->dndx = NUM_ENT(cursorp->pagep) - 1; - return (__ham_item(hashp, cursorp, mode)); + hcp->dndx = NUM_ENT(hcp->pagep) - 1; + return (__ham_item(dbc, mode)); } } @@ -362,95 +392,123 @@ __ham_item_prev(hashp, cursorp, mode) * midpage, beginning of page, beginning of bucket. */ - if (cursorp->bndx == 0) { /* Beginning of page. */ - if ((ret = __ham_get_cpage(hashp, cursorp, mode)) != 0) + if (F_ISSET(hcp, H_DUPONLY)) { + F_CLR(hcp, H_OK); + F_SET(hcp, H_NOMORE); + return (0); + } + + if (hcp->bndx == 0) { /* Beginning of page. */ + if ((ret = __ham_get_cpage(dbc, mode)) != 0) return (ret); - cursorp->pgno = PREV_PGNO(cursorp->pagep); - if (cursorp->pgno == PGNO_INVALID) { + hcp->pgno = PREV_PGNO(hcp->pagep); + if (hcp->pgno == PGNO_INVALID) { /* Beginning of bucket. */ - F_SET(cursorp, H_NOMORE); + F_SET(hcp, H_NOMORE); return (DB_NOTFOUND); - } else if ((ret = __ham_next_cpage(hashp, - cursorp, cursorp->pgno, 0, 0)) != 0) + } else if ((ret = + __ham_next_cpage(dbc, hcp->pgno, 0, 0)) != 0) return (ret); else - cursorp->bndx = H_NUMPAIRS(cursorp->pagep); + hcp->bndx = H_NUMPAIRS(hcp->pagep); } /* * Either we've got the cursor set up to be decremented, or we * have to find the end of a bucket. */ - if (cursorp->bndx == NDX_INVALID) { - if (cursorp->pagep == NULL) - next_pgno = BUCKET_TO_PAGE(hashp, cursorp->bucket); + if (hcp->bndx == NDX_INVALID) { + if (hcp->pagep == NULL) + next_pgno = BUCKET_TO_PAGE(hcp, hcp->bucket); else goto got_page; do { - if ((ret = __ham_next_cpage(hashp, - cursorp, next_pgno, 0, 0)) != 0) + if ((ret = __ham_next_cpage(dbc, next_pgno, 0, 0)) != 0) return (ret); -got_page: next_pgno = NEXT_PGNO(cursorp->pagep); - cursorp->bndx = H_NUMPAIRS(cursorp->pagep); +got_page: next_pgno = NEXT_PGNO(hcp->pagep); + hcp->bndx = H_NUMPAIRS(hcp->pagep); } while (next_pgno != PGNO_INVALID); - if (cursorp->bndx == 0) { + if (hcp->bndx == 0) { /* Bucket was empty. */ - F_SET(cursorp, H_NOMORE); + F_SET(hcp, H_NOMORE); return (DB_NOTFOUND); } } - cursorp->bndx--; + hcp->bndx--; - return (__ham_item(hashp, cursorp, mode)); + return (__ham_item(dbc, mode)); } /* * Sets the cursor to the next key/data pair on a page. * - * PUBLIC: int __ham_item_next __P((HTAB *, HASH_CURSOR *, db_lockmode_t)); + * PUBLIC: int __ham_item_next __P((DBC *, db_lockmode_t)); */ int -__ham_item_next(hashp, cursorp, mode) - HTAB *hashp; - HASH_CURSOR *cursorp; +__ham_item_next(dbc, mode) + DBC *dbc; db_lockmode_t mode; { + HASH_CURSOR *hcp; + + hcp = (HASH_CURSOR *)dbc->internal; /* * Deleted on-page duplicates are a weird case. If we delete the last * one, then our cursor is at the very end of a duplicate set and * we actually need to go on to the next key. */ - if (F_ISSET(cursorp, H_DELETED)) { - if (cursorp->bndx != NDX_INVALID && - F_ISSET(cursorp, H_ISDUP) && - cursorp->dpgno == PGNO_INVALID && - cursorp->dup_tlen == cursorp->dup_off) { - F_CLR(cursorp, H_ISDUP); - cursorp->dpgno = PGNO_INVALID; - cursorp->bndx++; + if (F_ISSET(hcp, H_DELETED)) { + if (hcp->bndx != NDX_INVALID && + F_ISSET(hcp, H_ISDUP) && + hcp->dpgno == PGNO_INVALID && + hcp->dup_tlen == hcp->dup_off) { + if (F_ISSET(hcp, H_DUPONLY)) { + F_CLR(hcp, H_OK); + F_SET(hcp, H_NOMORE); + return (0); + } else { + F_CLR(hcp, H_ISDUP); + hcp->dpgno = PGNO_INVALID; + hcp->bndx++; + } + } else if (!F_ISSET(hcp, H_ISDUP) && + F_ISSET(hcp, H_DUPONLY)) { + F_CLR(hcp, H_OK); + F_SET(hcp, H_NOMORE); + return (0); } - F_CLR(cursorp, H_DELETED); - } else if (cursorp->bndx == NDX_INVALID) { - cursorp->bndx = 0; - cursorp->dpgno = PGNO_INVALID; - F_CLR(cursorp, H_ISDUP); - } else if (F_ISSET(cursorp, H_ISDUP) && cursorp->dpgno != PGNO_INVALID) - cursorp->dndx++; - else if (F_ISSET(cursorp, H_ISDUP)) { - cursorp->dndx++; - cursorp->dup_off += DUP_SIZE(cursorp->dup_len); - if (cursorp->dup_off >= cursorp->dup_tlen) { - F_CLR(cursorp, H_ISDUP); - cursorp->dpgno = PGNO_INVALID; - cursorp->bndx++; + F_CLR(hcp, H_DELETED); + } else if (hcp->bndx == NDX_INVALID) { + hcp->bndx = 0; + hcp->dpgno = PGNO_INVALID; + F_CLR(hcp, H_ISDUP); + } else if (F_ISSET(hcp, H_ISDUP) && hcp->dpgno != PGNO_INVALID) + hcp->dndx++; + else if (F_ISSET(hcp, H_ISDUP)) { + if (hcp->dup_off + DUP_SIZE(hcp->dup_len) >= + hcp->dup_tlen && F_ISSET(hcp, H_DUPONLY)) { + F_CLR(hcp, H_OK); + F_SET(hcp, H_NOMORE); + return (0); + } + hcp->dndx++; + hcp->dup_off += DUP_SIZE(hcp->dup_len); + if (hcp->dup_off >= hcp->dup_tlen) { + F_CLR(hcp, H_ISDUP); + hcp->dpgno = PGNO_INVALID; + hcp->bndx++; } + } else if (F_ISSET(hcp, H_DUPONLY)) { + F_CLR(hcp, H_OK); + F_SET(hcp, H_NOMORE); + return (0); } else - cursorp->bndx++; + hcp->bndx++; - return (__ham_item(hashp, cursorp, mode)); + return (__ham_item(dbc, mode)); } /* @@ -537,18 +595,15 @@ __ham_reputpair(p, psize, ndx, key, data) /* - * PUBLIC: int __ham_del_pair __P((HTAB *, HASH_CURSOR *, int)); - * - * XXX - * TODO: if the item is an offdup, delete the other pages and then remove - * the pair. If the offpage page is 0, then you can just remove the pair. + * PUBLIC: int __ham_del_pair __P((DBC *, int)); */ int -__ham_del_pair(hashp, cursorp, reclaim_page) - HTAB *hashp; - HASH_CURSOR *cursorp; +__ham_del_pair(dbc, reclaim_page) + DBC *dbc; int reclaim_page; { + DB *dbp; + HASH_CURSOR *hcp; DBT data_dbt, key_dbt; DB_ENV *dbenv; DB_LSN new_lsn, *n_lsn, tmp_lsn; @@ -557,13 +612,16 @@ __ham_del_pair(hashp, cursorp, reclaim_page) db_pgno_t chg_pgno, pgno; int ret, tret; - dbenv = hashp->dbp->dbenv; - ndx = cursorp->bndx; - if (cursorp->pagep == NULL && (ret = - __ham_get_page(hashp->dbp, cursorp->pgno, &cursorp->pagep)) != 0) + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; + + dbenv = dbp->dbenv; + ndx = hcp->bndx; + if (hcp->pagep == NULL && + (ret = __ham_get_page(dbp, hcp->pgno, &hcp->pagep)) != 0) return (ret); - p = cursorp->pagep; + p = hcp->pagep; /* * We optimize for the normal case which is when neither the key nor @@ -576,7 +634,7 @@ __ham_del_pair(hashp, cursorp, reclaim_page) if (HPAGE_PTYPE(H_PAIRKEY(p, ndx)) == H_OFFPAGE) { memcpy(&pgno, HOFFPAGE_PGNO(P_ENTRY(p, H_KEYINDEX(ndx))), sizeof(db_pgno_t)); - ret = __db_doff(hashp->dbp, pgno, __ham_del_page); + ret = __db_doff(dbc, pgno, __ham_del_page); } if (ret == 0) @@ -585,14 +643,14 @@ __ham_del_pair(hashp, cursorp, reclaim_page) memcpy(&pgno, HOFFPAGE_PGNO(P_ENTRY(p, H_DATAINDEX(ndx))), sizeof(db_pgno_t)); - ret = __db_doff(hashp->dbp, pgno, __ham_del_page); + ret = __db_doff(dbc, pgno, __ham_del_page); break; case H_OFFDUP: memcpy(&pgno, HOFFDUP_PGNO(P_ENTRY(p, H_DATAINDEX(ndx))), sizeof(db_pgno_t)); - ret = __db_ddup(hashp->dbp, pgno, __ham_del_page); - F_CLR(cursorp, H_ISDUP); + ret = __db_ddup(dbc, pgno, __ham_del_page); + F_CLR(hcp, H_ISDUP); break; case H_DUPLICATE: /* @@ -600,7 +658,7 @@ __ham_del_pair(hashp, cursorp, reclaim_page) * we had better clear the flag so that we update the * cursor appropriately. */ - F_CLR(cursorp, H_ISDUP); + F_CLR(hcp, H_ISDUP); break; } @@ -608,17 +666,17 @@ __ham_del_pair(hashp, cursorp, reclaim_page) return (ret); /* Now log the delete off this page. */ - if (DB_LOGGING(hashp->dbp)) { + if (DB_LOGGING(dbc)) { key_dbt.data = P_ENTRY(p, H_KEYINDEX(ndx)); key_dbt.size = - LEN_HITEM(p, hashp->hdr->pagesize, H_KEYINDEX(ndx)); + LEN_HITEM(p, hcp->hdr->pagesize, H_KEYINDEX(ndx)); data_dbt.data = P_ENTRY(p, H_DATAINDEX(ndx)); data_dbt.size = - LEN_HITEM(p, hashp->hdr->pagesize, H_DATAINDEX(ndx)); + LEN_HITEM(p, hcp->hdr->pagesize, H_DATAINDEX(ndx)); if ((ret = __ham_insdel_log(dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, DELPAIR, - hashp->dbp->log_fileid, PGNO(p), (u_int32_t)ndx, + dbc->txn, &new_lsn, 0, DELPAIR, + dbp->log_fileid, PGNO(p), (u_int32_t)ndx, &LSN(p), &key_dbt, &data_dbt)) != 0) return (ret); @@ -626,15 +684,16 @@ __ham_del_pair(hashp, cursorp, reclaim_page) LSN(p) = new_lsn; } - __ham_dpair(hashp->dbp, p, ndx); + __ham_dpair(dbp, p, ndx); /* - * If we are locking, we will not maintain this. - * XXXX perhaps we can retain incremental numbers and apply them + * If we are locking, we will not maintain this, because it is + * a hot spot. + * XXX perhaps we can retain incremental numbers and apply them * later. */ - if (!F_ISSET(hashp->dbp, DB_AM_LOCKING)) - --hashp->hdr->nelem; + if (!F_ISSET(dbp, DB_AM_LOCKING)) + --hcp->hdr->nelem; /* * If we need to reclaim the page, then check if the page is empty. @@ -653,25 +712,25 @@ __ham_del_pair(hashp, cursorp, reclaim_page) * are more pages in the chain. */ if ((ret = - __ham_get_page(hashp->dbp, NEXT_PGNO(p), &n_pagep)) != 0) + __ham_get_page(dbp, NEXT_PGNO(p), &n_pagep)) != 0) return (ret); if (NEXT_PGNO(n_pagep) != PGNO_INVALID) { if ((ret = - __ham_get_page(hashp->dbp, NEXT_PGNO(n_pagep), + __ham_get_page(dbp, NEXT_PGNO(n_pagep), &nn_pagep)) != 0) { - (void) __ham_put_page(hashp->dbp, n_pagep, 0); + (void) __ham_put_page(dbp, n_pagep, 0); return (ret); } } - if (DB_LOGGING(hashp->dbp)) { + if (DB_LOGGING(dbc)) { key_dbt.data = n_pagep; - key_dbt.size = hashp->hdr->pagesize; + key_dbt.size = hcp->hdr->pagesize; if ((ret = __ham_copypage_log(dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, - hashp->dbp->log_fileid, PGNO(p), &LSN(p), - PGNO(n_pagep), &LSN(n_pagep), NEXT_PGNO(n_pagep), + dbc->txn, &new_lsn, 0, dbp->log_fileid, PGNO(p), + &LSN(p), PGNO(n_pagep), &LSN(n_pagep), + NEXT_PGNO(n_pagep), NEXT_PGNO(n_pagep) == PGNO_INVALID ? NULL : &LSN(nn_pagep), &key_dbt)) != 0) return (ret); @@ -684,12 +743,12 @@ __ham_del_pair(hashp, cursorp, reclaim_page) } if (NEXT_PGNO(n_pagep) != PGNO_INVALID) { PREV_PGNO(nn_pagep) = PGNO(p); - (void)__ham_put_page(hashp->dbp, nn_pagep, 1); + (void)__ham_put_page(dbp, nn_pagep, 1); } tmp_pgno = PGNO(p); tmp_lsn = LSN(p); - memcpy(p, n_pagep, hashp->hdr->pagesize); + memcpy(p, n_pagep, hcp->hdr->pagesize); PGNO(p) = tmp_pgno; LSN(p) = tmp_lsn; PREV_PGNO(p) = PGNO_INVALID; @@ -697,25 +756,25 @@ __ham_del_pair(hashp, cursorp, reclaim_page) /* * Cursor is advanced to the beginning of the next page. */ - cursorp->bndx = 0; - cursorp->pgno = PGNO(p); - F_SET(cursorp, H_DELETED); + hcp->bndx = 0; + hcp->pgno = PGNO(p); + F_SET(hcp, H_DELETED); chg_pgno = PGNO(p); - if ((ret = __ham_dirty_page(hashp, p)) != 0 || - (ret = __ham_del_page(hashp->dbp, n_pagep)) != 0) + if ((ret = __ham_dirty_page(dbp, p)) != 0 || + (ret = __ham_del_page(dbc, n_pagep)) != 0) return (ret); } else if (reclaim_page && NUM_ENT(p) == 0 && PREV_PGNO(p) != PGNO_INVALID) { PAGE *n_pagep, *p_pagep; if ((ret = - __ham_get_page(hashp->dbp, PREV_PGNO(p), &p_pagep)) != 0) + __ham_get_page(dbp, PREV_PGNO(p), &p_pagep)) != 0) return (ret); if (NEXT_PGNO(p) != PGNO_INVALID) { - if ((ret = __ham_get_page(hashp->dbp, + if ((ret = __ham_get_page(dbp, NEXT_PGNO(p), &n_pagep)) != 0) { - (void)__ham_put_page(hashp->dbp, p_pagep, 0); + (void)__ham_put_page(dbp, p_pagep, 0); return (ret); } n_lsn = &LSN(n_pagep); @@ -728,10 +787,10 @@ __ham_del_pair(hashp, cursorp, reclaim_page) if (n_pagep != NULL) PREV_PGNO(n_pagep) = PGNO(p_pagep); - if (DB_LOGGING(hashp->dbp)) { + if (DB_LOGGING(dbc)) { if ((ret = __ham_newpage_log(dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, DELOVFL, - hashp->dbp->log_fileid, PREV_PGNO(p), &LSN(p_pagep), + dbc->txn, &new_lsn, 0, DELOVFL, + dbp->log_fileid, PREV_PGNO(p), &LSN(p_pagep), PGNO(p), &LSN(p), NEXT_PGNO(p), n_lsn)) != 0) return (ret); @@ -741,21 +800,21 @@ __ham_del_pair(hashp, cursorp, reclaim_page) LSN(n_pagep) = new_lsn; LSN(p) = new_lsn; } - cursorp->pgno = NEXT_PGNO(p); - cursorp->bndx = 0; + hcp->pgno = NEXT_PGNO(p); + hcp->bndx = 0; /* * Since we are about to delete the cursor page and we have * just moved the cursor, we need to make sure that the * old page pointer isn't left hanging around in the cursor. */ - cursorp->pagep = NULL; + hcp->pagep = NULL; chg_pgno = PGNO(p); - ret = __ham_del_page(hashp->dbp, p); - if ((tret = __ham_put_page(hashp->dbp, p_pagep, 1)) != 0 && + ret = __ham_del_page(dbc, p); + if ((tret = __ham_put_page(dbp, p_pagep, 1)) != 0 && ret == 0) ret = tret; if (n_pagep != NULL && - (tret = __ham_put_page(hashp->dbp, n_pagep, 1)) != 0 && + (tret = __ham_put_page(dbp, n_pagep, 1)) != 0 && ret == 0) ret = tret; if (ret != 0) @@ -766,19 +825,19 @@ __ham_del_pair(hashp, cursorp, reclaim_page) * so that we update the cursor correctly on the next call * to next. */ - F_SET(cursorp, H_DELETED); - chg_pgno = cursorp->pgno; - ret = __ham_dirty_page(hashp, p); + F_SET(hcp, H_DELETED); + chg_pgno = hcp->pgno; + ret = __ham_dirty_page(dbp, p); } - __ham_c_update(cursorp, chg_pgno, 0, 0, 0); + __ham_c_update(hcp, chg_pgno, 0, 0, 0); /* * Since we just deleted a pair from the master page, anything - * in cursorp->dpgno should be cleared. + * in hcp->dpgno should be cleared. */ - cursorp->dpgno = PGNO_INVALID; + hcp->dpgno = PGNO_INVALID; - F_CLR(cursorp, H_OK); + F_CLR(hcp, H_OK); return (ret); } @@ -787,15 +846,16 @@ __ham_del_pair(hashp, cursorp, reclaim_page) * Given the key data indicated by the cursor, replace part/all of it * according to the fields in the dbt. * - * PUBLIC: int __ham_replpair __P((HTAB *, HASH_CURSOR *, DBT *, u_int32_t)); + * PUBLIC: int __ham_replpair __P((DBC *, DBT *, u_int32_t)); */ int -__ham_replpair(hashp, hcp, dbt, make_dup) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_replpair(dbc, dbt, make_dup) + DBC *dbc; DBT *dbt; u_int32_t make_dup; { + DB *dbp; + HASH_CURSOR *hcp; DBT old_dbt, tdata, tmp; DB_LSN new_lsn; int32_t change; /* XXX: Possible overflow. */ @@ -814,6 +874,8 @@ __ham_replpair(hashp, hcp, dbt, make_dup) * be the common case). We handle case 3 as a delete and * add. */ + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; /* * We need to compute the number of bytes that we are adding or @@ -833,7 +895,7 @@ __ham_replpair(hashp, hcp, dbt, make_dup) memcpy(&len, HOFFPAGE_TLEN(hk), sizeof(u_int32_t)); else len = LEN_HKEYDATA(hcp->pagep, - hashp->dbp->pgsize, H_DATAINDEX(hcp->bndx)); + dbp->pgsize, H_DATAINDEX(hcp->bndx)); if (dbt->doff + dbt->dlen > len) change += dbt->doff + dbt->dlen - len; @@ -854,41 +916,39 @@ __ham_replpair(hashp, hcp, dbt, make_dup) tmp.flags = 0; F_SET(&tmp, DB_DBT_MALLOC | DB_DBT_INTERNAL); if ((ret = - __db_ret(hashp->dbp, hcp->pagep, H_KEYINDEX(hcp->bndx), - &tmp, &hcp->big_key, &hcp->big_keylen)) != 0) + __db_ret(dbp, hcp->pagep, H_KEYINDEX(hcp->bndx), + &tmp, &dbc->rkey.data, &dbc->rkey.size)) != 0) return (ret); if (dbt->doff == 0 && dbt->dlen == len) { - ret = __ham_del_pair(hashp, hcp, 0); + ret = __ham_del_pair(dbc, 0); if (ret == 0) - ret = __ham_add_el(hashp, - hcp, &tmp, dbt, H_KEYDATA); + ret = __ham_add_el(dbc, &tmp, dbt, H_KEYDATA); } else { /* Case B */ type = HPAGE_PTYPE(hk) != H_OFFPAGE ? HPAGE_PTYPE(hk) : H_KEYDATA; tdata.flags = 0; F_SET(&tdata, DB_DBT_MALLOC | DB_DBT_INTERNAL); - if ((ret = __db_ret(hashp->dbp, hcp->pagep, - H_DATAINDEX(hcp->bndx), &tdata, &hcp->big_data, - &hcp->big_datalen)) != 0) + if ((ret = __db_ret(dbp, hcp->pagep, + H_DATAINDEX(hcp->bndx), &tdata, &dbc->rdata.data, + &dbc->rdata.size)) != 0) goto err; /* Now we can delete the item. */ - if ((ret = __ham_del_pair(hashp, hcp, 0)) != 0) { - __db_free(tdata.data); + if ((ret = __ham_del_pair(dbc, 0)) != 0) { + __os_free(tdata.data, tdata.size); goto err; } /* Now shift old data around to make room for new. */ if (change > 0) { - tdata.data = (void *)__db_realloc(tdata.data, - tdata.size + change); + if ((ret = __os_realloc(&tdata.data, + tdata.size + change)) != 0) + return (ret); memset((u_int8_t *)tdata.data + tdata.size, 0, change); } - if (tdata.data == NULL) - return (ENOMEM); end = (u_int8_t *)tdata.data + tdata.size; src = (u_int8_t *)tdata.data + dbt->doff + dbt->dlen; @@ -902,10 +962,10 @@ __ham_replpair(hashp, hcp, dbt, make_dup) tdata.size += change; /* Now add the pair. */ - ret = __ham_add_el(hashp, hcp, &tmp, &tdata, type); - __db_free(tdata.data); + ret = __ham_add_el(dbc, &tmp, &tdata, type); + __os_free(tdata.data, tdata.size); } -err: __db_free(tmp.data); +err: __os_free(tmp.data, tmp.size); return (ret); } @@ -921,12 +981,11 @@ err: __db_free(tmp.data); * all the parameters here. Then log the call before moving * anything around. */ - if (DB_LOGGING(hashp->dbp)) { + if (DB_LOGGING(dbc)) { old_dbt.data = beg; old_dbt.size = dbt->dlen; - if ((ret = __ham_replace_log(hashp->dbp->dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, - hashp->dbp->log_fileid, PGNO(hcp->pagep), + if ((ret = __ham_replace_log(dbp->dbenv->lg_info, + dbc->txn, &new_lsn, 0, dbp->log_fileid, PGNO(hcp->pagep), (u_int32_t)H_DATAINDEX(hcp->bndx), &LSN(hcp->pagep), (u_int32_t)dbt->doff, &old_dbt, dbt, make_dup)) != 0) return (ret); @@ -934,7 +993,7 @@ err: __db_free(tmp.data); LSN(hcp->pagep) = new_lsn; /* Structure assignment. */ } - __ham_onpage_replace(hcp->pagep, hashp->dbp->pgsize, + __ham_onpage_replace(hcp->pagep, dbp->pgsize, (u_int32_t)H_DATAINDEX(hcp->bndx), (int32_t)dbt->doff, change, dbt); return (0); @@ -997,13 +1056,15 @@ __ham_onpage_replace(pagep, pgsize, ndx, off, change, dbt) } /* - * PUBLIC: int __ham_split_page __P((HTAB *, u_int32_t, u_int32_t)); + * PUBLIC: int __ham_split_page __P((DBC *, u_int32_t, u_int32_t)); */ int -__ham_split_page(hashp, obucket, nbucket) - HTAB *hashp; +__ham_split_page(dbc, obucket, nbucket) + DBC *dbc; u_int32_t obucket, nbucket; { + DB *dbp; + HASH_CURSOR *hcp; DBT key, page_dbt; DB_ENV *dbenv; DB_LSN new_lsn; @@ -1014,33 +1075,34 @@ __ham_split_page(hashp, obucket, nbucket) int ret, tret; void *big_buf; - dbenv = hashp->dbp->dbenv; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; + dbenv = dbp->dbenv; temp_pagep = old_pagep = new_pagep = NULL; - bucket_pgno = BUCKET_TO_PAGE(hashp, obucket); - if ((ret = __ham_get_page(hashp->dbp, bucket_pgno, &old_pagep)) != 0) + bucket_pgno = BUCKET_TO_PAGE(hcp, obucket); + if ((ret = __ham_get_page(dbp, bucket_pgno, &old_pagep)) != 0) return (ret); - if ((ret = __ham_new_page(hashp, BUCKET_TO_PAGE(hashp, nbucket), P_HASH, + if ((ret = __ham_new_page(dbp, BUCKET_TO_PAGE(hcp, nbucket), P_HASH, &new_pagep)) != 0) goto err; - temp_pagep = hashp->split_buf; - memcpy(temp_pagep, old_pagep, hashp->hdr->pagesize); + temp_pagep = hcp->split_buf; + memcpy(temp_pagep, old_pagep, hcp->hdr->pagesize); - if (DB_LOGGING(hashp->dbp)) { - page_dbt.size = hashp->hdr->pagesize; + if (DB_LOGGING(dbc)) { + page_dbt.size = hcp->hdr->pagesize; page_dbt.data = old_pagep; if ((ret = __ham_splitdata_log(dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, - hashp->dbp->log_fileid, SPLITOLD, PGNO(old_pagep), - &page_dbt, &LSN(old_pagep))) != 0) + dbc->txn, &new_lsn, 0, dbp->log_fileid, SPLITOLD, + PGNO(old_pagep), &page_dbt, &LSN(old_pagep))) != 0) goto err; } - P_INIT(old_pagep, hashp->hdr->pagesize, PGNO(old_pagep), PGNO_INVALID, + P_INIT(old_pagep, hcp->hdr->pagesize, PGNO(old_pagep), PGNO_INVALID, PGNO_INVALID, 0, P_HASH); - if (DB_LOGGING(hashp->dbp)) + if (DB_LOGGING(dbc)) LSN(old_pagep) = new_lsn; /* Structure assignment. */ big_len = 0; @@ -1049,11 +1111,11 @@ __ham_split_page(hashp, obucket, nbucket) while (temp_pagep != NULL) { for (n = 0; n < (db_indx_t)H_NUMPAIRS(temp_pagep); n++) { if ((ret = - __db_ret(hashp->dbp, temp_pagep, H_KEYINDEX(n), + __db_ret(dbp, temp_pagep, H_KEYINDEX(n), &key, &big_buf, &big_len)) != 0) goto err; - if (__ham_call_hash(hashp, key.data, key.size) + if (__ham_call_hash(hcp, key.data, key.size) == obucket) pp = &old_pagep; else @@ -1064,59 +1126,59 @@ __ham_split_page(hashp, obucket, nbucket) * page to store the key/data pair. */ - len = LEN_HITEM(temp_pagep, hashp->hdr->pagesize, + len = LEN_HITEM(temp_pagep, hcp->hdr->pagesize, H_DATAINDEX(n)) + - LEN_HITEM(temp_pagep, hashp->hdr->pagesize, + LEN_HITEM(temp_pagep, hcp->hdr->pagesize, H_KEYINDEX(n)) + 2 * sizeof(db_indx_t); if (P_FREESPACE(*pp) < len) { - if (DB_LOGGING(hashp->dbp)) { - page_dbt.size = hashp->hdr->pagesize; + if (DB_LOGGING(dbc)) { + page_dbt.size = hcp->hdr->pagesize; page_dbt.data = *pp; if ((ret = __ham_splitdata_log( - dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, - &new_lsn, 0, - hashp->dbp->log_fileid, SPLITNEW, - PGNO(*pp), &page_dbt, + dbenv->lg_info, dbc->txn, + &new_lsn, 0, dbp->log_fileid, + SPLITNEW, PGNO(*pp), &page_dbt, &LSN(*pp))) != 0) goto err; LSN(*pp) = new_lsn; } - if ((ret = __ham_add_ovflpage(hashp, - *pp, 1, pp)) != 0) + if ((ret = + __ham_add_ovflpage(dbc, *pp, 1, pp)) != 0) goto err; } - __ham_copy_item(hashp, temp_pagep, H_KEYINDEX(n), *pp); - __ham_copy_item(hashp, temp_pagep, H_DATAINDEX(n), *pp); + __ham_copy_item(dbp->pgsize, + temp_pagep, H_KEYINDEX(n), *pp); + __ham_copy_item(dbp->pgsize, + temp_pagep, H_DATAINDEX(n), *pp); } next_pgno = NEXT_PGNO(temp_pagep); /* Clear temp_page; if it's a link overflow page, free it. */ if (PGNO(temp_pagep) != bucket_pgno && (ret = - __ham_del_page(hashp->dbp, temp_pagep)) != 0) + __ham_del_page(dbc, temp_pagep)) != 0) goto err; if (next_pgno == PGNO_INVALID) temp_pagep = NULL; else if ((ret = - __ham_get_page(hashp->dbp, next_pgno, &temp_pagep)) != 0) + __ham_get_page(dbp, next_pgno, &temp_pagep)) != 0) goto err; - if (temp_pagep != NULL && DB_LOGGING(hashp->dbp)) { - page_dbt.size = hashp->hdr->pagesize; + if (temp_pagep != NULL && DB_LOGGING(dbc)) { + page_dbt.size = hcp->hdr->pagesize; page_dbt.data = temp_pagep; if ((ret = __ham_splitdata_log(dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, - hashp->dbp->log_fileid, SPLITOLD, PGNO(temp_pagep), + dbc->txn, &new_lsn, 0, dbp->log_fileid, + SPLITOLD, PGNO(temp_pagep), &page_dbt, &LSN(temp_pagep))) != 0) goto err; LSN(temp_pagep) = new_lsn; } } if (big_buf != NULL) - __db_free(big_buf); + __os_free(big_buf, big_len); /* * If the original bucket spanned multiple pages, then we've got @@ -1124,42 +1186,41 @@ __ham_split_page(hashp, obucket, nbucket) * should be deleted. */ if (temp_pagep != NULL && PGNO(temp_pagep) != bucket_pgno && - (ret = __ham_del_page(hashp->dbp, temp_pagep)) != 0) + (ret = __ham_del_page(dbc, temp_pagep)) != 0) goto err; /* * Write new buckets out. */ - if (DB_LOGGING(hashp->dbp)) { - page_dbt.size = hashp->hdr->pagesize; + if (DB_LOGGING(dbc)) { + page_dbt.size = hcp->hdr->pagesize; page_dbt.data = old_pagep; if ((ret = __ham_splitdata_log(dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, - hashp->dbp->log_fileid, SPLITNEW, PGNO(old_pagep), + dbc->txn, &new_lsn, 0, dbp->log_fileid, + SPLITNEW, PGNO(old_pagep), &page_dbt, &LSN(old_pagep))) != 0) goto err; LSN(old_pagep) = new_lsn; page_dbt.data = new_pagep; if ((ret = __ham_splitdata_log(dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, - hashp->dbp->log_fileid, SPLITNEW, PGNO(new_pagep), - &page_dbt, &LSN(new_pagep))) != 0) + dbc->txn, &new_lsn, 0, dbp->log_fileid, + SPLITNEW, PGNO(new_pagep), &page_dbt, &LSN(new_pagep))) != 0) goto err; LSN(new_pagep) = new_lsn; } - ret = __ham_put_page(hashp->dbp, old_pagep, 1); - if ((tret = __ham_put_page(hashp->dbp, new_pagep, 1)) != 0 && + ret = __ham_put_page(dbp, old_pagep, 1); + if ((tret = __ham_put_page(dbp, new_pagep, 1)) != 0 && ret == 0) ret = tret; if (0) { err: if (old_pagep != NULL) - (void)__ham_put_page(hashp->dbp, old_pagep, 1); + (void)__ham_put_page(dbp, old_pagep, 1); if (new_pagep != NULL) - (void)__ham_put_page(hashp->dbp, new_pagep, 1); + (void)__ham_put_page(dbp, new_pagep, 1); if (temp_pagep != NULL && PGNO(temp_pagep) != bucket_pgno) - (void)__ham_put_page(hashp->dbp, temp_pagep, 1); + (void)__ham_put_page(dbp, temp_pagep, 1); } return (ret); } @@ -1171,16 +1232,16 @@ err: if (old_pagep != NULL) * to which we just added something. This allows us to link overflow * pages and return the new page having correctly put the last page. * - * PUBLIC: int __ham_add_el - * PUBLIC: __P((HTAB *, HASH_CURSOR *, const DBT *, const DBT *, int)); + * PUBLIC: int __ham_add_el __P((DBC *, const DBT *, const DBT *, int)); */ int -__ham_add_el(hashp, hcp, key, val, type) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_add_el(dbc, key, val, type) + DBC *dbc; const DBT *key, *val; int type; { + DB *dbp; + HASH_CURSOR *hcp; const DBT *pkey, *pdata; DBT key_dbt, data_dbt; DB_LSN new_lsn; @@ -1190,17 +1251,19 @@ __ham_add_el(hashp, hcp, key, val, type) int do_expand, is_keybig, is_databig, ret; int key_type, data_type; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; do_expand = 0; - if (hcp->pagep == NULL && (ret = __ham_get_page(hashp->dbp, + if (hcp->pagep == NULL && (ret = __ham_get_page(dbp, hcp->seek_found_page != PGNO_INVALID ? hcp->seek_found_page : hcp->pgno, &hcp->pagep)) != 0) return (ret); key_size = HKEYDATA_PSIZE(key->size); data_size = HKEYDATA_PSIZE(val->size); - is_keybig = ISBIG(hashp, key->size); - is_databig = ISBIG(hashp, val->size); + is_keybig = ISBIG(hcp, key->size); + is_databig = ISBIG(hcp, val->size); if (is_keybig) key_size = HOFFPAGE_PSIZE; if (is_databig) @@ -1220,7 +1283,7 @@ __ham_add_el(hashp, hcp, key, val, type) break; next_pgno = NEXT_PGNO(hcp->pagep); if ((ret = - __ham_next_cpage(hashp, hcp, next_pgno, 0, 0)) != 0) + __ham_next_cpage(dbc, next_pgno, 0, 0)) != 0) return (ret); } @@ -1229,7 +1292,7 @@ __ham_add_el(hashp, hcp, key, val, type) */ if (P_FREESPACE(hcp->pagep) < pairsize) { do_expand = 1; - if ((ret = __ham_add_ovflpage(hashp, + if ((ret = __ham_add_ovflpage(dbc, hcp->pagep, 1, &hcp->pagep)) != 0) return (ret); hcp->pgno = PGNO(hcp->pagep); @@ -1241,10 +1304,13 @@ __ham_add_el(hashp, hcp, key, val, type) hcp->bndx = H_NUMPAIRS(hcp->pagep); F_CLR(hcp, H_DELETED); if (is_keybig) { - if ((ret = __db_poff(hashp->dbp, + koff.type = H_OFFPAGE; + UMRW(koff.unused[0]); + UMRW(koff.unused[1]); + UMRW(koff.unused[2]); + if ((ret = __db_poff(dbc, key, &koff.pgno, __ham_overflow_page)) != 0) return (ret); - koff.type = H_OFFPAGE; koff.tlen = key->size; key_dbt.data = &koff; key_dbt.size = sizeof(koff); @@ -1256,10 +1322,13 @@ __ham_add_el(hashp, hcp, key, val, type) } if (is_databig) { - if ((ret = __db_poff(hashp->dbp, + doff.type = H_OFFPAGE; + UMRW(doff.unused[0]); + UMRW(doff.unused[1]); + UMRW(doff.unused[2]); + if ((ret = __db_poff(dbc, val, &doff.pgno, __ham_overflow_page)) != 0) return (ret); - doff.type = H_OFFPAGE; doff.tlen = val->size; data_dbt.data = &doff; data_dbt.size = sizeof(doff); @@ -1270,16 +1339,16 @@ __ham_add_el(hashp, hcp, key, val, type) data_type = type; } - if (DB_LOGGING(hashp->dbp)) { + if (DB_LOGGING(dbc)) { rectype = PUTPAIR; if (is_databig) rectype |= PAIR_DATAMASK; if (is_keybig) rectype |= PAIR_KEYMASK; - if ((ret = __ham_insdel_log(hashp->dbp->dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, rectype, - hashp->dbp->log_fileid, PGNO(hcp->pagep), + if ((ret = __ham_insdel_log(dbp->dbenv->lg_info, + dbc->txn, &new_lsn, 0, rectype, + dbp->log_fileid, PGNO(hcp->pagep), (u_int32_t)H_NUMPAIRS(hcp->pagep), &LSN(hcp->pagep), pkey, pdata)) != 0) return (ret); @@ -1303,11 +1372,11 @@ __ham_add_el(hashp, hcp, key, val, type) /* * XXX Maybe keep incremental numbers here */ - if (!F_ISSET(hashp->dbp, DB_AM_LOCKING)) - hashp->hdr->nelem++; + if (!F_ISSET(dbp, DB_AM_LOCKING)) + hcp->hdr->nelem++; - if (do_expand || (hashp->hdr->ffactor != 0 && - (u_int32_t)H_NUMPAIRS(hcp->pagep) > hashp->hdr->ffactor)) + if (do_expand || (hcp->hdr->ffactor != 0 && + (u_int32_t)H_NUMPAIRS(hcp->pagep) > hcp->hdr->ffactor)) F_SET(hcp, H_EXPAND); return (0); } @@ -1319,11 +1388,11 @@ __ham_add_el(hashp, hcp, key, val, type) * H_DUPLICATE, H_OFFDUP). Since we log splits at a high level, we * do not need to do any logging here. * - * PUBLIC: void __ham_copy_item __P((HTAB *, PAGE *, u_int32_t, PAGE *)); + * PUBLIC: void __ham_copy_item __P((size_t, PAGE *, u_int32_t, PAGE *)); */ void -__ham_copy_item(hashp, src_page, src_ndx, dest_page) - HTAB *hashp; +__ham_copy_item(pgsize, src_page, src_ndx, dest_page) + size_t pgsize; PAGE *src_page; u_int32_t src_ndx; PAGE *dest_page; @@ -1337,7 +1406,7 @@ __ham_copy_item(hashp, src_page, src_ndx, dest_page) src = P_ENTRY(src_page, src_ndx); /* Set up space on dest. */ - len = LEN_HITEM(src_page, hashp->hdr->pagesize, src_ndx); + len = LEN_HITEM(src_page, pgsize, src_ndx); HOFFSET(dest_page) -= len; dest_page->inp[NUM_ENT(dest_page)] = HOFFSET(dest_page); dest = P_ENTRY(dest_page, NUM_ENT(dest_page)); @@ -1352,29 +1421,31 @@ __ham_copy_item(hashp, src_page, src_ndx, dest_page) * pointer on success * NULL on error * - * PUBLIC: int __ham_add_ovflpage __P((HTAB *, PAGE *, int, PAGE **)); + * PUBLIC: int __ham_add_ovflpage __P((DBC *, PAGE *, int, PAGE **)); */ int -__ham_add_ovflpage(hashp, pagep, release, pp) - HTAB *hashp; +__ham_add_ovflpage(dbc, pagep, release, pp) + DBC *dbc; PAGE *pagep; int release; PAGE **pp; { - DB_ENV *dbenv; + DB *dbp; + HASH_CURSOR *hcp; DB_LSN new_lsn; PAGE *new_pagep; int ret; - dbenv = hashp->dbp->dbenv; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; - if ((ret = __ham_overflow_page(hashp->dbp, P_HASH, &new_pagep)) != 0) + if ((ret = __ham_overflow_page(dbc, P_HASH, &new_pagep)) != 0) return (ret); - if (DB_LOGGING(hashp->dbp)) { - if ((ret = __ham_newpage_log(dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, PUTOVFL, - hashp->dbp->log_fileid, PGNO(pagep), &LSN(pagep), + if (DB_LOGGING(dbc)) { + if ((ret = __ham_newpage_log(dbp->dbenv->lg_info, + dbc->txn, &new_lsn, 0, PUTOVFL, + dbp->log_fileid, PGNO(pagep), &LSN(pagep), PGNO(new_pagep), &LSN(new_pagep), PGNO_INVALID, NULL)) != 0) return (ret); @@ -1385,78 +1456,76 @@ __ham_add_ovflpage(hashp, pagep, release, pp) PREV_PGNO(new_pagep) = PGNO(pagep); if (release) - ret = __ham_put_page(hashp->dbp, pagep, 1); + ret = __ham_put_page(dbp, pagep, 1); - hashp->hash_overflows++; + hcp->stats.hash_overflows++; *pp = new_pagep; return (ret); } /* - * PUBLIC: int __ham_new_page __P((HTAB *, u_int32_t, u_int32_t, PAGE **)); + * PUBLIC: int __ham_new_page __P((DB *, u_int32_t, u_int32_t, PAGE **)); */ int -__ham_new_page(hashp, addr, type, pp) - HTAB *hashp; +__ham_new_page(dbp, addr, type, pp) + DB *dbp; u_int32_t addr, type; PAGE **pp; { PAGE *pagep; int ret; - if ((ret = memp_fget(hashp->dbp->mpf, + if ((ret = memp_fget(dbp->mpf, &addr, DB_MPOOL_CREATE, &pagep)) != 0) return (ret); -#ifdef DEBUG_SLOW - __account_page(hashp, addr, 1); -#endif /* This should not be necessary because page-in should do it. */ - P_INIT(pagep, - hashp->hdr->pagesize, addr, PGNO_INVALID, PGNO_INVALID, 0, type); + P_INIT(pagep, dbp->pgsize, addr, PGNO_INVALID, PGNO_INVALID, 0, type); *pp = pagep; return (0); } /* - * PUBLIC: int __ham_del_page __P((DB *, PAGE *)); + * PUBLIC: int __ham_del_page __P((DBC *, PAGE *)); */ int -__ham_del_page(dbp, pagep) - DB *dbp; +__ham_del_page(dbc, pagep) + DBC *dbc; PAGE *pagep; { + DB *dbp; + HASH_CURSOR *hcp; DB_LSN new_lsn; - HTAB *hashp; int ret; - hashp = (HTAB *)dbp->internal; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; ret = 0; - DIRTY_META(hashp, ret); + DIRTY_META(dbp, hcp, ret); if (ret != 0) { if (ret != EAGAIN) - __db_err(hashp->dbp->dbenv, + __db_err(dbp->dbenv, "free_ovflpage: unable to lock meta data page %s\n", strerror(ret)); /* * If we are going to return an error, then we should free * the page, so it doesn't stay pinned forever. */ - (void)__ham_put_page(hashp->dbp, pagep, 0); + (void)__ham_put_page(dbp, pagep, 0); return (ret); } - if (DB_LOGGING(hashp->dbp)) { - if ((ret = __ham_newpgno_log(hashp->dbp->dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, DELPGNO, - hashp->dbp->log_fileid, PGNO(pagep), hashp->hdr->last_freed, + if (DB_LOGGING(dbc)) { + if ((ret = __ham_newpgno_log(dbp->dbenv->lg_info, + dbc->txn, &new_lsn, 0, DELPGNO, + dbp->log_fileid, PGNO(pagep), hcp->hdr->last_freed, (u_int32_t)TYPE(pagep), NEXT_PGNO(pagep), P_INVALID, - &LSN(pagep), &hashp->hdr->lsn)) != 0) + &LSN(pagep), &hcp->hdr->lsn)) != 0) return (ret); - hashp->hdr->lsn = new_lsn; + hcp->hdr->lsn = new_lsn; LSN(pagep) = new_lsn; } @@ -1466,16 +1535,16 @@ __ham_del_page(dbp, pagep) DB_LSN __lsn; __pgno = pagep->pgno; __lsn = pagep->lsn; - memset(pagep, 0xff, dbp->pgsize); + memset(pagep, 0xdb, dbp->pgsize); pagep->pgno = __pgno; pagep->lsn = __lsn; } #endif TYPE(pagep) = P_INVALID; - NEXT_PGNO(pagep) = hashp->hdr->last_freed; - hashp->hdr->last_freed = PGNO(pagep); + NEXT_PGNO(pagep) = hcp->hdr->last_freed; + hcp->hdr->last_freed = PGNO(pagep); - return (__ham_put_page(hashp->dbp, pagep, 1)); + return (__ham_put_page(dbp, pagep, 1)); } @@ -1489,8 +1558,7 @@ __ham_put_page(dbp, pagep, is_dirty) int32_t is_dirty; { #ifdef DEBUG_SLOW - __account_page((HTAB *)dbp->cookie, - ((BKT *)((char *)pagep - sizeof(BKT)))->pgno, -1); + __account_page(dbp, ((BKT *)((char *)pagep - sizeof(BKT)))->pgno, -1); #endif return (memp_fput(dbp->mpf, pagep, (is_dirty ? DB_MPOOL_DIRTY : 0))); } @@ -1499,14 +1567,14 @@ __ham_put_page(dbp, pagep, is_dirty) * __ham_dirty_page -- * Mark a page dirty. * - * PUBLIC: int __ham_dirty_page __P((HTAB *, PAGE *)); + * PUBLIC: int __ham_dirty_page __P((DB *, PAGE *)); */ int -__ham_dirty_page(hashp, pagep) - HTAB *hashp; +__ham_dirty_page(dbp, pagep) + DB *dbp; PAGE *pagep; { - return (memp_fset(hashp->dbp->mpf, pagep, DB_MPOOL_DIRTY)); + return (memp_fset(dbp->mpf, pagep, DB_MPOOL_DIRTY)); } /* @@ -1523,31 +1591,33 @@ __ham_get_page(dbp, addr, pagep) ret = memp_fget(dbp->mpf, &addr, DB_MPOOL_CREATE, pagep); #ifdef DEBUG_SLOW if (*pagep != NULL) - __account_page((HTAB *)dbp->internal, addr, 1); + __account_page(dbp, addr, 1); #endif return (ret); } /* - * PUBLIC: int __ham_overflow_page __P((DB *, u_int32_t, PAGE **)); + * PUBLIC: int __ham_overflow_page + * PUBLIC: __P((DBC *, u_int32_t, PAGE **)); */ int -__ham_overflow_page(dbp, type, pp) - DB *dbp; +__ham_overflow_page(dbc, type, pp) + DBC *dbc; u_int32_t type; PAGE **pp; { + DB *dbp; + HASH_CURSOR *hcp; DB_LSN *lsnp, new_lsn; - HTAB *hashp; PAGE *p; db_pgno_t new_addr, next_free, newalloc_flag; u_int32_t offset, splitnum; int ret; - hashp = (HTAB *)dbp->internal; - ret = 0; - DIRTY_META(hashp, ret); + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; + DIRTY_META(dbp, hcp, ret); if (ret != 0) return (ret); @@ -1558,22 +1628,22 @@ __ham_overflow_page(dbp, type, pp) * after the log do we get to complete allocation of the * new page. */ - new_addr = hashp->hdr->last_freed; + new_addr = hcp->hdr->last_freed; if (new_addr != PGNO_INVALID) { - if ((ret = __ham_get_page(hashp->dbp, new_addr, &p)) != 0) + if ((ret = __ham_get_page(dbp, new_addr, &p)) != 0) return (ret); next_free = NEXT_PGNO(p); lsnp = &LSN(p); newalloc_flag = 0; } else { - splitnum = hashp->hdr->ovfl_point; - hashp->hdr->spares[splitnum]++; - offset = hashp->hdr->spares[splitnum] - - (splitnum ? hashp->hdr->spares[splitnum - 1] : 0); - new_addr = PGNO_OF(hashp, hashp->hdr->ovfl_point, offset); - if (new_addr > MAX_PAGES(hashp)) { - __db_err(hashp->dbp->dbenv, "hash: out of file pages"); - hashp->hdr->spares[splitnum]--; + splitnum = hcp->hdr->ovfl_point; + hcp->hdr->spares[splitnum]++; + offset = hcp->hdr->spares[splitnum] - + (splitnum ? hcp->hdr->spares[splitnum - 1] : 0); + new_addr = PGNO_OF(hcp, hcp->hdr->ovfl_point, offset); + if (new_addr > MAX_PAGES(hcp)) { + __db_err(dbp->dbenv, "hash: out of file pages"); + hcp->hdr->spares[splitnum]--; return (ENOMEM); } next_free = PGNO_INVALID; @@ -1582,29 +1652,29 @@ __ham_overflow_page(dbp, type, pp) newalloc_flag = 1; } - if (DB_LOGGING(hashp->dbp)) { - if ((ret = __ham_newpgno_log(hashp->dbp->dbenv->lg_info, - (DB_TXN *)hashp->dbp->txn, &new_lsn, 0, ALLOCPGNO, - hashp->dbp->log_fileid, new_addr, next_free, - 0, newalloc_flag, type, lsnp, &hashp->hdr->lsn)) != 0) + if (DB_LOGGING(dbc)) { + if ((ret = __ham_newpgno_log(dbp->dbenv->lg_info, + dbc->txn, &new_lsn, 0, ALLOCPGNO, + dbp->log_fileid, new_addr, next_free, + 0, newalloc_flag, type, lsnp, &hcp->hdr->lsn)) != 0) return (ret); - hashp->hdr->lsn = new_lsn; + hcp->hdr->lsn = new_lsn; if (lsnp != NULL) *lsnp = new_lsn; } if (p != NULL) { /* We just took something off the free list, initialize it. */ - hashp->hdr->last_freed = next_free; - P_INIT(p, hashp->hdr->pagesize, PGNO(p), PGNO_INVALID, + hcp->hdr->last_freed = next_free; + P_INIT(p, hcp->hdr->pagesize, PGNO(p), PGNO_INVALID, PGNO_INVALID, 0, (u_int8_t)type); } else { /* Get the new page. */ - if ((ret = __ham_new_page(hashp, new_addr, type, &p)) != 0) + if ((ret = __ham_new_page(dbp, new_addr, type, &p)) != 0) return (ret); } - if (DB_LOGGING(hashp->dbp)) + if (DB_LOGGING(dbc)) LSN(p) = new_lsn; *pp = p; @@ -1614,94 +1684,123 @@ __ham_overflow_page(dbp, type, pp) #ifdef DEBUG /* * PUBLIC: #ifdef DEBUG - * PUBLIC: db_pgno_t __bucket_to_page __P((HTAB *, db_pgno_t)); + * PUBLIC: db_pgno_t __bucket_to_page __P((HASH_CURSOR *, db_pgno_t)); * PUBLIC: #endif */ db_pgno_t -__bucket_to_page(hashp, n) - HTAB *hashp; +__bucket_to_page(hcp, n) + HASH_CURSOR *hcp; db_pgno_t n; { int ret_val; ret_val = n + 1; if (n != 0) - ret_val += hashp->hdr->spares[__db_log2(n + 1) - 1]; + ret_val += hcp->hdr->spares[__db_log2(n + 1) - 1]; return (ret_val); } #endif /* * Create a bunch of overflow pages at the current split point. - * PUBLIC: void __ham_init_ovflpages __P((HTAB *)); + * PUBLIC: void __ham_init_ovflpages __P((DBC *)); */ void -__ham_init_ovflpages(hp) - HTAB *hp; +__ham_init_ovflpages(dbc) + DBC *dbc; { + DB *dbp; + HASH_CURSOR *hcp; DB_LSN new_lsn; PAGE *p; db_pgno_t last_pgno, new_pgno; u_int32_t i, curpages, numpages; - curpages = hp->hdr->spares[hp->hdr->ovfl_point] - - hp->hdr->spares[hp->hdr->ovfl_point - 1]; - numpages = hp->hdr->ovfl_point + 1 - curpages; - - last_pgno = hp->hdr->last_freed; - new_pgno = PGNO_OF(hp, hp->hdr->ovfl_point, curpages + 1); - if (DB_LOGGING(hp->dbp)) { - (void)__ham_ovfl_log(hp->dbp->dbenv->lg_info, - (DB_TXN *)hp->dbp->txn, &new_lsn, 0, - hp->dbp->log_fileid, new_pgno, - numpages, last_pgno, hp->hdr->ovfl_point, &hp->hdr->lsn); - hp->hdr->lsn = new_lsn; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; + + curpages = hcp->hdr->spares[hcp->hdr->ovfl_point] - + hcp->hdr->spares[hcp->hdr->ovfl_point - 1]; + numpages = hcp->hdr->ovfl_point + 1 - curpages; + + last_pgno = hcp->hdr->last_freed; + new_pgno = PGNO_OF(hcp, hcp->hdr->ovfl_point, curpages + 1); + if (DB_LOGGING(dbc)) { + (void)__ham_ovfl_log(dbp->dbenv->lg_info, + dbc->txn, &new_lsn, 0, dbp->log_fileid, new_pgno, + numpages, last_pgno, hcp->hdr->ovfl_point, &hcp->hdr->lsn); + hcp->hdr->lsn = new_lsn; } else ZERO_LSN(new_lsn); - hp->hdr->spares[hp->hdr->ovfl_point] += numpages; + hcp->hdr->spares[hcp->hdr->ovfl_point] += numpages; for (i = numpages; i > 0; i--) { - if (__ham_new_page(hp, - PGNO_OF(hp, hp->hdr->ovfl_point, curpages + i), + if (__ham_new_page(dbp, + PGNO_OF(hcp, hcp->hdr->ovfl_point, curpages + i), P_INVALID, &p) != 0) break; LSN(p) = new_lsn; NEXT_PGNO(p) = last_pgno; last_pgno = PGNO(p); - (void)__ham_put_page(hp->dbp, p, 1); + (void)__ham_put_page(dbp, p, 1); } - hp->hdr->last_freed = last_pgno; + hcp->hdr->last_freed = last_pgno; } /* - * PUBLIC: int __ham_get_cpage __P((HTAB *, HASH_CURSOR *, db_lockmode_t)); + * PUBLIC: int __ham_get_cpage __P((DBC *, db_lockmode_t)); */ int -__ham_get_cpage(hashp, hcp, mode) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_get_cpage(dbc, mode) + DBC *dbc; db_lockmode_t mode; { + DB *dbp; + HASH_CURSOR *hcp; int ret; - if (hcp->lock == 0 && F_ISSET(hashp->dbp, DB_AM_LOCKING) && - (ret = __ham_lock_bucket(hashp->dbp, hcp, mode)) != 0) - return (ret); + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; + + /* + * There are three cases with respect to buckets and locks. If there + * is no lock held, then if we are locking, we should get the lock. + * If there is a lock held and it's for the current bucket, we don't + * need to do anything. If there is a lock, but it's for a different + * bucket, then we need to release and get. + */ + if (F_ISSET(dbp, DB_AM_LOCKING)) { + if (hcp->lock != 0 && hcp->lbucket != hcp->bucket) { + /* + * If this is the original lock, don't release it, + * because we may need to restore it upon exit. + */ + if (dbc->txn == NULL && + !F_ISSET(hcp, H_ORIGINAL) && (ret = + lock_put(dbp->dbenv->lk_info, hcp->lock)) != 0) + return (ret); + F_CLR(hcp, H_ORIGINAL); + hcp->lock = 0; + } + if (hcp->lock == 0 && (ret = __ham_lock_bucket(dbc, mode)) != 0) + return (ret); + hcp->lbucket = hcp->bucket; + } if (hcp->pagep == NULL) { if (hcp->pgno == PGNO_INVALID) { - hcp->pgno = BUCKET_TO_PAGE(hashp, hcp->bucket); + hcp->pgno = BUCKET_TO_PAGE(hcp, hcp->bucket); hcp->bndx = 0; } if ((ret = - __ham_get_page(hashp->dbp, hcp->pgno, &hcp->pagep)) != 0) + __ham_get_page(dbp, hcp->pgno, &hcp->pagep)) != 0) return (ret); } if (hcp->dpgno != PGNO_INVALID && hcp->dpagep == NULL) if ((ret = - __ham_get_page(hashp->dbp, hcp->dpgno, &hcp->dpagep)) != 0) + __ham_get_page(dbp, hcp->dpgno, &hcp->dpagep)) != 0) return (ret); return (0); } @@ -1711,28 +1810,30 @@ __ham_get_cpage(hashp, hcp, mode) * If the flag is set to H_ISDUP, then we are talking about the * duplicate page, not the main page. * - * PUBLIC: int __ham_next_cpage - * PUBLIC: __P((HTAB *, HASH_CURSOR *, db_pgno_t, int, u_int32_t)); + * PUBLIC: int __ham_next_cpage __P((DBC *, db_pgno_t, int, u_int32_t)); */ int -__ham_next_cpage(hashp, hcp, pgno, dirty, flags) - HTAB *hashp; - HASH_CURSOR *hcp; +__ham_next_cpage(dbc, pgno, dirty, flags) + DBC *dbc; db_pgno_t pgno; int dirty; u_int32_t flags; { + DB *dbp; + HASH_CURSOR *hcp; PAGE *p; int ret; + dbp = dbc->dbp; + hcp = (HASH_CURSOR *)dbc->internal; if (LF_ISSET(H_ISDUP) && hcp->dpagep != NULL && - (ret = __ham_put_page(hashp->dbp, hcp->dpagep, dirty)) != 0) + (ret = __ham_put_page(dbp, hcp->dpagep, dirty)) != 0) return (ret); else if (!LF_ISSET(H_ISDUP) && hcp->pagep != NULL && - (ret = __ham_put_page(hashp->dbp, hcp->pagep, dirty)) != 0) + (ret = __ham_put_page(dbp, hcp->pagep, dirty)) != 0) return (ret); - if ((ret = __ham_get_page(hashp->dbp, pgno, &p)) != 0) + if ((ret = __ham_get_page(dbp, pgno, &p)) != 0) return (ret); if (LF_ISSET(H_ISDUP)) { @@ -1753,22 +1854,21 @@ __ham_next_cpage(hashp, hcp, pgno, dirty, flags) * Get the lock on a particular bucket. */ static int -__ham_lock_bucket(dbp, hcp, mode) - DB *dbp; - HASH_CURSOR *hcp; +__ham_lock_bucket(dbc, mode) + DBC *dbc; db_lockmode_t mode; { + HASH_CURSOR *hcp; int ret; - /* - * What a way to trounce on the memory system. It might be - * worth copying the lk_info into the hashp. - */ - ret = 0; - dbp->lock.pgno = (db_pgno_t)(hcp->bucket); - ret = lock_get(dbp->dbenv->lk_info, - dbp->txn == NULL ? dbp->locker : dbp->txn->txnid, 0, - &dbp->lock_dbt, mode, &hcp->lock); + hcp = (HASH_CURSOR *)dbc->internal; + dbc->lock.pgno = (db_pgno_t)(hcp->bucket); + if (dbc->txn == NULL) + ret = lock_get(dbc->dbp->dbenv->lk_info, dbc->locker, 0, + &dbc->lock_dbt, mode, &hcp->lock); + else + ret = lock_tget(dbc->dbp->dbenv->lk_info, dbc->txn, 0, + &dbc->lock_dbt, mode, &hcp->lock); return (ret < 0 ? EAGAIN : ret); } @@ -1827,45 +1927,3 @@ __ham_dpair(dbp, p, pndx) HOFFSET(p) = HOFFSET(p) + delta; NUM_ENT(p) = NUM_ENT(p) - 2; } - -#ifdef DEBUG_SLOW -static void -__account_page(hashp, pgno, inout) - HTAB *hashp; - db_pgno_t pgno; - int inout; -{ - static struct { - db_pgno_t pgno; - int times; - } list[100]; - static int last; - int i, j; - - if (inout == -1) /* XXX: Kluge */ - inout = 0; - - /* Find page in list. */ - for (i = 0; i < last; i++) - if (list[i].pgno == pgno) - break; - /* Not found. */ - if (i == last) { - list[last].times = inout; - list[last].pgno = pgno; - last++; - } - list[i].times = inout; - if (list[i].times == 0) { - for (j = i; j < last; j++) - list[j] = list[j + 1]; - last--; - } - for (i = 0; i < last; i++, list[i].times++) - if (list[i].times > 20 && - !__is_bitmap_pgno(hashp, list[i].pgno)) - (void)fprintf(stderr, - "Warning: pg %lu has been out for %d times\n", - (u_long)list[i].pgno, list[i].times); -} -#endif /* DEBUG_SLOW */ diff --git a/db2/hash/hash_rec.c b/db2/hash/hash_rec.c index 727f615828..b58f2c6eb7 100644 --- a/db2/hash/hash_rec.c +++ b/db2/hash/hash_rec.c @@ -47,7 +47,7 @@ #include "config.h" #ifndef lint -static const char sccsid[] = "@(#)hash_rec.c 10.19 (Sleepycat) 5/23/98"; +static const char sccsid[] = "@(#)hash_rec.c 10.22 (Sleepycat) 10/21/98"; #endif /* not lint */ #ifndef NO_SYSTEM_INCLUDES @@ -80,17 +80,19 @@ __ham_insdel_recover(logp, dbtp, lsnp, redo, info) void *info; { __ham_insdel_args *argp; - DB *mdbp, *file_dbp; + DB *file_dbp; + DBC *dbc; + HASH_CURSOR *hcp; DB_MPOOLFILE *mpf; - HTAB *hashp; PAGE *pagep; u_int32_t op; int cmp_n, cmp_p, getmeta, ret; getmeta = 0; - hashp = NULL; /* XXX: shut the compiler up. */ + hcp = NULL; REC_PRINT(__ham_insdel_print); REC_INTRO(__ham_insdel_read); + hcp = (HASH_CURSOR *)dbc->internal; ret = memp_fget(mpf, &argp->pgno, 0, &pagep); if (ret != 0) { @@ -101,16 +103,15 @@ __ham_insdel_recover(logp, dbtp, lsnp, redo, info) * would not have to undo anything. In this case, * don't bother creating a page. */ - *lsnp = argp->prev_lsn; - ret = 0; - goto out; + goto done; } else if ((ret = memp_fget(mpf, &argp->pgno, DB_MPOOL_CREATE, &pagep)) != 0) goto out; } - hashp = (HTAB *)file_dbp->internal; - GET_META(file_dbp, hashp); + GET_META(file_dbp, hcp, ret); + if (ret != 0) + goto out; getmeta = 1; cmp_n = log_compare(lsnp, &LSN(pagep)); @@ -144,7 +145,7 @@ __ham_insdel_recover(logp, dbtp, lsnp, redo, info) !redo || PAIR_ISDATABIG(argp->opcode) ? H_OFFPAGE : H_KEYDATA); } else - (void) __ham_reputpair(pagep, hashp->hdr->pagesize, + (void) __ham_reputpair(pagep, hcp->hdr->pagesize, argp->ndx, &argp->key, &argp->data); LSN(pagep) = redo ? *lsnp : argp->pagelsn; @@ -163,10 +164,11 @@ __ham_insdel_recover(logp, dbtp, lsnp, redo, info) goto out; /* Return the previous LSN. */ - *lsnp = argp->prev_lsn; +done: *lsnp = argp->prev_lsn; + ret = 0; out: if (getmeta) - RELEASE_META(file_dbp, hashp); + RELEASE_META(file_dbp, hcp); REC_CLOSE; } @@ -187,16 +189,18 @@ __ham_newpage_recover(logp, dbtp, lsnp, redo, info) void *info; { __ham_newpage_args *argp; - DB *mdbp, *file_dbp; + DB *file_dbp; + DBC *dbc; + HASH_CURSOR *hcp; DB_MPOOLFILE *mpf; - HTAB *hashp; PAGE *pagep; int cmp_n, cmp_p, change, getmeta, ret; getmeta = 0; - hashp = NULL; /* XXX: shut the compiler up. */ + hcp = NULL; REC_PRINT(__ham_newpage_print); REC_INTRO(__ham_newpage_read); + hcp = (HASH_CURSOR *)dbc->internal; ret = memp_fget(mpf, &argp->new_pgno, 0, &pagep); if (ret != 0) { @@ -214,8 +218,9 @@ __ham_newpage_recover(logp, dbtp, lsnp, redo, info) goto out; } - hashp = (HTAB *)file_dbp->internal; - GET_META(file_dbp, hashp); + GET_META(file_dbp, (HASH_CURSOR *)dbc->internal, ret); + if (ret != 0) + goto out; getmeta = 1; /* @@ -289,11 +294,13 @@ ppage: if (argp->prev_pgno != PGNO_INVALID) { } if (!change) { - if ((ret = __ham_put_page(file_dbp, (PAGE *)pagep, 0)) != 0) + if ((ret = + __ham_put_page(file_dbp, (PAGE *)pagep, 0)) != 0) goto out; } else { LSN(pagep) = redo ? *lsnp : argp->prevlsn; - if ((ret = __ham_put_page(file_dbp, (PAGE *)pagep, 1)) != 0) + if ((ret = + __ham_put_page(file_dbp, (PAGE *)pagep, 1)) != 0) goto out; } } @@ -310,9 +317,7 @@ npage: if (argp->next_pgno != PGNO_INVALID) { * so we would not have to undo anything. In * this case, don't bother creating a page. */ - *lsnp = argp->prev_lsn; - ret = 0; - goto out; + goto done; } else if ((ret = memp_fget(mpf, &argp->next_pgno, DB_MPOOL_CREATE, &pagep)) != 0) @@ -346,10 +351,11 @@ npage: if (argp->next_pgno != PGNO_INVALID) { goto out; } } - *lsnp = argp->prev_lsn; +done: *lsnp = argp->prev_lsn; + ret = 0; out: if (getmeta) - RELEASE_META(file_dbp, hashp); + RELEASE_META(file_dbp, hcp); REC_CLOSE; } @@ -372,19 +378,21 @@ __ham_replace_recover(logp, dbtp, lsnp, redo, info) void *info; { __ham_replace_args *argp; - DB *mdbp, *file_dbp; + DB *file_dbp; + DBC *dbc; + HASH_CURSOR *hcp; DB_MPOOLFILE *mpf; DBT dbt; - HTAB *hashp; PAGE *pagep; int32_t grow; int change, cmp_n, cmp_p, getmeta, ret; u_int8_t *hk; getmeta = 0; - hashp = NULL; /* XXX: shut the compiler up. */ + hcp = NULL; REC_PRINT(__ham_replace_print); REC_INTRO(__ham_replace_read); + hcp = (HASH_CURSOR *)dbc->internal; ret = memp_fget(mpf, &argp->pgno, 0, &pagep); if (ret != 0) { @@ -395,16 +403,15 @@ __ham_replace_recover(logp, dbtp, lsnp, redo, info) * would not have to undo anything. In this case, * don't bother creating a page. */ - *lsnp = argp->prev_lsn; - ret = 0; - goto out; + goto done; } else if ((ret = memp_fget(mpf, &argp->pgno, DB_MPOOL_CREATE, &pagep)) != 0) goto out; } - hashp = (HTAB *)file_dbp->internal; - GET_META(file_dbp, hashp); + GET_META(file_dbp, (HASH_CURSOR *)dbc->internal, ret); + if (ret != 0) + goto out; getmeta = 1; cmp_n = log_compare(lsnp, &LSN(pagep)); @@ -444,10 +451,11 @@ __ham_replace_recover(logp, dbtp, lsnp, redo, info) if ((ret = __ham_put_page(file_dbp, pagep, change)) != 0) goto out; - *lsnp = argp->prev_lsn; +done: *lsnp = argp->prev_lsn; + ret = 0; out: if (getmeta) - RELEASE_META(file_dbp, hashp); + RELEASE_META(file_dbp, hcp); REC_CLOSE; } @@ -468,19 +476,22 @@ __ham_newpgno_recover(logp, dbtp, lsnp, redo, info) void *info; { __ham_newpgno_args *argp; - DB *mdbp, *file_dbp; + DB *file_dbp; + DBC *dbc; + HASH_CURSOR *hcp; DB_MPOOLFILE *mpf; - HTAB *hashp; PAGE *pagep; int change, cmp_n, cmp_p, getmeta, ret; getmeta = 0; - hashp = NULL; /* XXX: shut the compiler up. */ + hcp = NULL; REC_PRINT(__ham_newpgno_print); REC_INTRO(__ham_newpgno_read); + hcp = (HASH_CURSOR *)dbc->internal; - hashp = (HTAB *)file_dbp->internal; - GET_META(file_dbp, hashp); + GET_META(file_dbp, (HASH_CURSOR *)dbc->internal, ret); + if (ret != 0) + goto out; getmeta = 1; /* @@ -488,34 +499,34 @@ __ham_newpgno_recover(logp, dbtp, lsnp, redo, info) * to update the meta data; then we need to update the page. * We'll do the meta-data first. */ - cmp_n = log_compare(lsnp, &hashp->hdr->lsn); - cmp_p = log_compare(&hashp->hdr->lsn, &argp->metalsn); + cmp_n = log_compare(lsnp, &hcp->hdr->lsn); + cmp_p = log_compare(&hcp->hdr->lsn, &argp->metalsn); change = 0; if ((cmp_p == 0 && redo && argp->opcode == ALLOCPGNO) || (cmp_n == 0 && !redo && argp->opcode == DELPGNO)) { /* Need to redo an allocation or undo a deletion. */ - hashp->hdr->last_freed = argp->free_pgno; + hcp->hdr->last_freed = argp->free_pgno; if (redo && argp->old_pgno != 0) /* Must be ALLOCPGNO */ - hashp->hdr->spares[hashp->hdr->ovfl_point]++; + hcp->hdr->spares[hcp->hdr->ovfl_point]++; change = 1; } else if (cmp_p == 0 && redo && argp->opcode == DELPGNO) { /* Need to redo a deletion */ - hashp->hdr->last_freed = argp->pgno; + hcp->hdr->last_freed = argp->pgno; change = 1; } else if (cmp_n == 0 && !redo && argp->opcode == ALLOCPGNO) { /* undo an allocation. */ if (argp->old_pgno == 0) - hashp->hdr->last_freed = argp->pgno; + hcp->hdr->last_freed = argp->pgno; else { - hashp->hdr->spares[hashp->hdr->ovfl_point]--; - hashp->hdr->last_freed = 0; + hcp->hdr->spares[hcp->hdr->ovfl_point]--; + hcp->hdr->last_freed = 0; } change = 1; } if (change) { - hashp->hdr->lsn = redo ? *lsnp : argp->metalsn; - F_SET(file_dbp, DB_HS_DIRTYMETA); + hcp->hdr->lsn = redo ? *lsnp : argp->metalsn; + F_SET(hcp, H_DIRTY); } @@ -530,9 +541,7 @@ __ham_newpgno_recover(logp, dbtp, lsnp, redo, info) * would not have to undo anything. In this case, * don't bother creating a page. */ - *lsnp = argp->prev_lsn; - ret = 0; - goto out; + goto done; } else if ((ret = memp_fget(mpf, &argp->pgno, DB_MPOOL_CREATE, &pagep)) != 0) goto out; @@ -565,10 +574,11 @@ __ham_newpgno_recover(logp, dbtp, lsnp, redo, info) if ((ret = __ham_put_page(file_dbp, pagep, change)) != 0) goto out; - *lsnp = argp->prev_lsn; +done: *lsnp = argp->prev_lsn; + ret = 0; out: if (getmeta) - RELEASE_META(file_dbp, hashp); + RELEASE_META(file_dbp, hcp); REC_CLOSE; } @@ -590,19 +600,22 @@ __ham_splitmeta_recover(logp, dbtp, lsnp, redo, info) void *info; { __ham_splitmeta_args *argp; - DB *mdbp, *file_dbp; + DB *file_dbp; + DBC *dbc; + HASH_CURSOR *hcp; DB_MPOOLFILE *mpf; - HTAB *hashp; int change, cmp_n, cmp_p, getmeta, ret; u_int32_t pow; getmeta = 0; - hashp = NULL; /* XXX: shut the compiler up. */ + hcp = NULL; REC_PRINT(__ham_splitmeta_print); REC_INTRO(__ham_splitmeta_read); + hcp = (HASH_CURSOR *)dbc->internal; - hashp = (HTAB *)file_dbp->internal; - GET_META(file_dbp, hashp); + GET_META(file_dbp, (HASH_CURSOR *)dbc->internal, ret); + if (ret != 0) + goto out; getmeta = 1; /* @@ -610,43 +623,45 @@ __ham_splitmeta_recover(logp, dbtp, lsnp, redo, info) * to update the meta data; then we need to update the page. * We'll do the meta-data first. */ - cmp_n = log_compare(lsnp, &hashp->hdr->lsn); - cmp_p = log_compare(&hashp->hdr->lsn, &argp->metalsn); + cmp_n = log_compare(lsnp, &hcp->hdr->lsn); + cmp_p = log_compare(&hcp->hdr->lsn, &argp->metalsn); change = 0; if (cmp_p == 0 && redo) { /* Need to redo the split information. */ - hashp->hdr->max_bucket = argp->bucket + 1; - pow = __db_log2(hashp->hdr->max_bucket + 1); - if (pow > hashp->hdr->ovfl_point) { - hashp->hdr->spares[pow] = - hashp->hdr->spares[hashp->hdr->ovfl_point]; - hashp->hdr->ovfl_point = pow; + hcp->hdr->max_bucket = argp->bucket + 1; + pow = __db_log2(hcp->hdr->max_bucket + 1); + if (pow > hcp->hdr->ovfl_point) { + hcp->hdr->spares[pow] = + hcp->hdr->spares[hcp->hdr->ovfl_point]; + hcp->hdr->ovfl_point = pow; } - if (hashp->hdr->max_bucket > hashp->hdr->high_mask) { - hashp->hdr->low_mask = hashp->hdr->high_mask; - hashp->hdr->high_mask = - hashp->hdr->max_bucket | hashp->hdr->low_mask; + if (hcp->hdr->max_bucket > hcp->hdr->high_mask) { + hcp->hdr->low_mask = hcp->hdr->high_mask; + hcp->hdr->high_mask = + hcp->hdr->max_bucket | hcp->hdr->low_mask; } change = 1; } else if (cmp_n == 0 && !redo) { /* Need to undo the split information. */ - hashp->hdr->max_bucket = argp->bucket; - hashp->hdr->ovfl_point = argp->ovflpoint; - hashp->hdr->spares[hashp->hdr->ovfl_point] = argp->spares; - pow = 1 << __db_log2(hashp->hdr->max_bucket + 1); - hashp->hdr->high_mask = pow - 1; - hashp->hdr->low_mask = (pow >> 1) - 1; + hcp->hdr->max_bucket = argp->bucket; + hcp->hdr->ovfl_point = argp->ovflpoint; + hcp->hdr->spares[hcp->hdr->ovfl_point] = argp->spares; + pow = 1 << __db_log2(hcp->hdr->max_bucket + 1); + hcp->hdr->high_mask = pow - 1; + hcp->hdr->low_mask = (pow >> 1) - 1; change = 1; } if (change) { - hashp->hdr->lsn = redo ? *lsnp : argp->metalsn; - F_SET(file_dbp, DB_HS_DIRTYMETA); + hcp->hdr->lsn = redo ? *lsnp : argp->metalsn; + F_SET(hcp, H_DIRTY); } - *lsnp = argp->prev_lsn; + +done: *lsnp = argp->prev_lsn; + ret = 0; out: if (getmeta) - RELEASE_META(file_dbp, hashp); + RELEASE_META(file_dbp, hcp); REC_CLOSE; } @@ -665,16 +680,18 @@ __ham_splitdata_recover(logp, dbtp, lsnp, redo, info) void *info; { __ham_splitdata_args *argp; - DB *mdbp, *file_dbp; + DB *file_dbp; + DBC *dbc; + HASH_CURSOR *hcp; DB_MPOOLFILE *mpf; - HTAB *hashp; PAGE *pagep; int change, cmp_n, cmp_p, getmeta, ret; getmeta = 0; - hashp = NULL; /* XXX: shut the compiler up. */ + hcp = NULL; REC_PRINT(__ham_splitdata_print); REC_INTRO(__ham_splitdata_read); + hcp = (HASH_CURSOR *)dbc->internal; ret = memp_fget(mpf, &argp->pgno, 0, &pagep); if (ret != 0) { @@ -685,16 +702,15 @@ __ham_splitdata_recover(logp, dbtp, lsnp, redo, info) * would not have to undo anything. In this case, * don't bother creating a page. */ - *lsnp = argp->prev_lsn; - ret = 0; - goto out; + goto done; } else if ((ret = memp_fget(mpf, &argp->pgno, DB_MPOOL_CREATE, &pagep)) != 0) goto out; } - hashp = (HTAB *)file_dbp->internal; - GET_META(file_dbp, hashp); + GET_META(file_dbp, (HASH_CURSOR *)dbc->internal, ret); + if (ret != 0) + goto out; getmeta = 1; cmp_n = log_compare(lsnp, &LSN(pagep)); @@ -732,10 +748,11 @@ __ham_splitdata_recover(logp, dbtp, lsnp, redo, info) if ((ret = __ham_put_page(file_dbp, pagep, change)) != 0) goto out; - *lsnp = argp->prev_lsn; +done: *lsnp = argp->prev_lsn; + ret = 0; out: if (getmeta) - RELEASE_META(file_dbp, hashp); + RELEASE_META(file_dbp, hcp); REC_CLOSE; } @@ -755,50 +772,52 @@ __ham_ovfl_recover(logp, dbtp, lsnp, redo, info) void *info; { __ham_ovfl_args *argp; - DB *mdbp, *file_dbp; + DB *file_dbp; + DBC *dbc; + HASH_CURSOR *hcp; DB_MPOOLFILE *mpf; - HTAB *hashp; PAGE *pagep; db_pgno_t max_pgno, pgno; int cmp_n, cmp_p, getmeta, ret; getmeta = 0; - hashp = NULL; /* XXX: shut the compiler up. */ + hcp = NULL; REC_PRINT(__ham_ovfl_print); REC_INTRO(__ham_ovfl_read); + hcp = (HASH_CURSOR *)dbc->internal; - hashp = (HTAB *)file_dbp->internal; - GET_META(file_dbp, hashp); + GET_META(file_dbp, (HASH_CURSOR *)dbc->internal, ret); + if (ret != 0) + goto out; getmeta = 1; - cmp_n = log_compare(lsnp, &hashp->hdr->lsn); - cmp_p = log_compare(&hashp->hdr->lsn, &argp->metalsn); + cmp_n = log_compare(lsnp, &hcp->hdr->lsn); + cmp_p = log_compare(&hcp->hdr->lsn, &argp->metalsn); if (cmp_p == 0 && redo) { /* Redo the allocation. */ - hashp->hdr->last_freed = argp->start_pgno; - hashp->hdr->spares[argp->ovflpoint] += argp->npages; - hashp->hdr->lsn = *lsnp; - F_SET(file_dbp, DB_HS_DIRTYMETA); + hcp->hdr->last_freed = argp->start_pgno; + hcp->hdr->spares[argp->ovflpoint] += argp->npages; + hcp->hdr->lsn = *lsnp; + F_SET(hcp, H_DIRTY); } else if (cmp_n == 0 && !redo) { - hashp->hdr->last_freed = argp->free_pgno; - hashp->hdr->spares[argp->ovflpoint] -= argp->npages; - hashp->hdr->lsn = argp->metalsn; - F_SET(file_dbp, DB_HS_DIRTYMETA); + hcp->hdr->last_freed = argp->free_pgno; + hcp->hdr->spares[argp->ovflpoint] -= argp->npages; + hcp->hdr->lsn = argp->metalsn; + F_SET(hcp, H_DIRTY); } max_pgno = argp->start_pgno + argp->npages - 1; ret = 0; for (pgno = argp->start_pgno; pgno <= max_pgno; pgno++) { - ret = memp_fget(mpf, &pgno, 0, &pagep); - if (ret != 0) { - if (redo && (ret = memp_fget(mpf, &pgno, - DB_MPOOL_CREATE, &pagep)) != 0) - goto out; - else if (!redo) { - (void)__ham_put_page(file_dbp, pagep, 0); + if ((ret = memp_fget(mpf, &pgno, 0, &pagep)) != 0) { + if (!redo) { + ret = 0; continue; } + if ((ret = memp_fget(mpf, + &pgno, DB_MPOOL_CREATE, &pagep)) != 0) + goto out; } if (redo && log_compare((const DB_LSN *)lsnp, (const DB_LSN *)&LSN(pagep)) > 0) { @@ -816,9 +835,11 @@ __ham_ovfl_recover(logp, dbtp, lsnp, redo, info) goto out; } - *lsnp = argp->prev_lsn; +done: *lsnp = argp->prev_lsn; + ret = 0; + out: if (getmeta) - RELEASE_META(file_dbp, hashp); + RELEASE_META(file_dbp, hcp); REC_CLOSE; } @@ -838,19 +859,22 @@ __ham_copypage_recover(logp, dbtp, lsnp, redo, info) void *info; { __ham_copypage_args *argp; - DB *file_dbp, *mdbp; + DB *file_dbp; + DBC *dbc; + HASH_CURSOR *hcp; DB_MPOOLFILE *mpf; - HTAB *hashp; PAGE *pagep; int cmp_n, cmp_p, getmeta, modified, ret; getmeta = 0; - hashp = NULL; /* XXX: shut the compiler up. */ + hcp = NULL; REC_PRINT(__ham_copypage_print); REC_INTRO(__ham_copypage_read); + hcp = (HASH_CURSOR *)dbc->internal; - hashp = (HTAB *)file_dbp->internal; - GET_META(file_dbp, hashp); + GET_META(file_dbp, (HASH_CURSOR *)dbc->internal, ret); + if (ret != 0) + goto out; getmeta = 1; modified = 0; @@ -881,7 +905,7 @@ __ham_copypage_recover(logp, dbtp, lsnp, redo, info) modified = 1; } else if (cmp_n == 0 && !redo) { /* Need to undo update described. */ - P_INIT(pagep, hashp->hdr->pagesize, argp->pgno, PGNO_INVALID, + P_INIT(pagep, hcp->hdr->pagesize, argp->pgno, PGNO_INVALID, argp->next_pgno, 0, P_HASH); LSN(pagep) = argp->pagelsn; modified = 1; @@ -918,10 +942,8 @@ donext: ret = memp_fget(mpf, &argp->next_pgno, 0, &pagep); goto out; /* Now fix up the next's next page. */ -do_nn: if (argp->nnext_pgno == PGNO_INVALID) { - *lsnp = argp->prev_lsn; - goto out; - } +do_nn: if (argp->nnext_pgno == PGNO_INVALID) + goto done; ret = memp_fget(mpf, &argp->nnext_pgno, 0, &pagep); if (ret != 0) { @@ -932,9 +954,7 @@ do_nn: if (argp->nnext_pgno == PGNO_INVALID) { * would not have to undo anything. In this case, * don't bother creating a page. */ - ret = 0; - *lsnp = argp->prev_lsn; - goto out; + goto done; } else if ((ret = memp_fget(mpf, &argp->nnext_pgno, DB_MPOOL_CREATE, &pagep)) != 0) goto out; @@ -957,9 +977,10 @@ do_nn: if (argp->nnext_pgno == PGNO_INVALID) { if ((ret = memp_fput(mpf, pagep, modified ? DB_MPOOL_DIRTY : 0)) != 0) goto out; - *lsnp = argp->prev_lsn; +done: *lsnp = argp->prev_lsn; + ret = 0; out: if (getmeta) - RELEASE_META(file_dbp, hashp); + RELEASE_META(file_dbp, hcp); REC_CLOSE; } diff --git a/db2/hash/hash_stat.c b/db2/hash/hash_stat.c index b57ca0950d..1b493d5f40 100644 --- a/db2/hash/hash_stat.c +++ b/db2/hash/hash_stat.c @@ -8,7 +8,7 @@ #include "config.h" #ifndef lint -static const char sccsid[] = "@(#)hash_stat.c 10.8 (Sleepycat) 4/26/98"; +static const char sccsid[] = "@(#)hash_stat.c 10.12 (Sleepycat) 12/19/98"; #endif /* not lint */ #ifndef NO_SYSTEM_INCLUDES @@ -23,35 +23,22 @@ static const char sccsid[] = "@(#)hash_stat.c 10.8 (Sleepycat) 4/26/98"; /* * __ham_stat -- - * Gather/print the hash statistics. + * Gather/print the hash statistics * - * PUBLIC: int __ham_stat __P((DB *, FILE *)); + * PUBLIC: int __ham_stat __P((DB *, void *, void *(*)(size_t), u_int32_t)); */ int -__ham_stat(dbp, fp) +__ham_stat(dbp, spp, db_malloc, flags) DB *dbp; - FILE *fp; + void *spp; + void *(*db_malloc) __P((size_t)); + u_int32_t flags; { - HTAB *hashp; - int i; + COMPQUIET(spp, NULL); + COMPQUIET(db_malloc, NULL); + COMPQUIET(flags, 0); - hashp = (HTAB *)dbp->internal; + DB_PANIC_CHECK(dbp); - fprintf(fp, "hash: accesses %lu collisions %lu\n", - hashp->hash_accesses, hashp->hash_collisions); - fprintf(fp, "hash: expansions %lu\n", hashp->hash_expansions); - fprintf(fp, "hash: overflows %lu\n", hashp->hash_overflows); - fprintf(fp, "hash: big key/data pages %lu\n", hashp->hash_bigpages); - - SET_LOCKER(dbp, NULL); - GET_META(dbp, hashp); - fprintf(fp, "keys %lu maxp %lu\n", - (u_long)hashp->hdr->nelem, (u_long)hashp->hdr->max_bucket); - - for (i = 0; i < NCACHED; i++) - fprintf(fp, - "spares[%d] = %lu\n", i, (u_long)hashp->hdr->spares[i]); - - RELEASE_META(dbp, hashp); - return (0); + return (__db_eopnotsup(dbp->dbenv)); } |