/*- * See the file LICENSE for redistribution information. * * Copyright (c) 1996-2001 * Sleepycat Software. All rights reserved. */ /* * Copyright (c) 1995, 1996 * The President and Fellows of Harvard University. All rights reserved. * * This code is derived from software contributed to Berkeley by * Margo Seltzer. * * 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. 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 "db_config.h" #ifndef lint static const char revid[] = "$Id: db_dispatch.c,v 11.75 2001/11/16 21:44:19 bostic Exp $"; #endif /* not lint */ #ifndef NO_SYSTEM_INCLUDES #include #include #include #include #endif #include "db_int.h" #include "db_page.h" #include "db_dispatch.h" #include "db_am.h" #include "log_auto.h" #include "txn.h" #include "txn_auto.h" #include "log.h" static int __db_txnlist_find_internal __P((DB_ENV *, void *, db_txnlist_type, u_int32_t, u_int8_t [DB_FILE_ID_LEN], DB_TXNLIST **, int)); static int __db_txnlist_pgnoadd __P((DB_ENV *, DB_TXNHEAD *, int32_t, u_int8_t [DB_FILE_ID_LEN], char *, db_pgno_t)); /* * __db_dispatch -- * * This is the transaction dispatch function used by the db access methods. * It is designed to handle the record format used by all the access * methods (the one automatically generated by the db_{h,log,read}.sh * scripts in the tools directory). An application using a different * recovery paradigm will supply a different dispatch function to txn_open. * * PUBLIC: int __db_dispatch __P((DB_ENV *, * PUBLIC: int (**)__P((DB_ENV *, DBT *, DB_LSN *, db_recops, void *)), * PUBLIC: DBT *, DB_LSN *, db_recops, void *)); */ int __db_dispatch(dbenv, dtab, db, lsnp, redo, info) DB_ENV *dbenv; /* The environment. */ int (**dtab)__P((DB_ENV *, DBT *, DB_LSN *, db_recops, void *)); DBT *db; /* The log record upon which to dispatch. */ DB_LSN *lsnp; /* The lsn of the record being dispatched. */ db_recops redo; /* Redo this op (or undo it). */ void *info; { DB_LSN prev_lsn; u_int32_t rectype, txnid; int make_call, ret; memcpy(&rectype, db->data, sizeof(rectype)); memcpy(&txnid, (u_int8_t *)db->data + sizeof(rectype), sizeof(txnid)); make_call = ret = 0; /* * If we find a record that is in the user's number space and they * have specified a recovery routine, let them handle it. If they * didn't specify a recovery routine, then we expect that they've * followed all our rules and registered new recovery functions. */ switch (redo) { case DB_TXN_ABORT: /* * XXX * db_printlog depends on DB_TXN_ABORT not examining the TXN * list. If that ever changes, fix db_printlog too. */ make_call = 1; break; case DB_TXN_OPENFILES: case DB_TXN_POPENFILES: if (rectype == DB_log_register || rectype == DB_txn_ckp) return (dbenv->dtab[rectype](dbenv, db, lsnp, redo, info)); else if (redo == DB_TXN_OPENFILES && txnid != 0) { /* * We collect all the transactions that have * "begin" records, those with no previous LSN, * so that we do not abort partial transactions. * These are known to be undone, otherwise the * log would not have been freeable. */ memcpy(&prev_lsn, (u_int8_t *)db->data + sizeof(rectype) + sizeof(txnid), sizeof(prev_lsn)); if (prev_lsn.file == 0) return (__db_txnlist_add( dbenv, info, txnid, TXN_OK, NULL)); } break; case DB_TXN_BACKWARD_ROLL: /* * Running full recovery in the backward pass. If we've * seen this txnid before and added to it our commit list, * then we do nothing during this pass, unless this is a child * commit record, in which case we need to process it. If * we've never seen it, then we call the appropriate recovery * routine. * * We need to always undo DB_db_noop records, so that we * properly handle any aborts before the file was closed. */ switch(rectype) { case DB_txn_regop: case DB_log_register: case DB_txn_ckp: case DB_db_noop: case DB_txn_child: make_call = 1; break; default: if (txnid != 0 && (ret = __db_txnlist_find(dbenv, info, txnid)) != TXN_COMMIT && ret != TXN_IGNORE) { /* * If not found then, this is an incomplete * abort. */ if (ret == TXN_NOTFOUND) return (__db_txnlist_add(dbenv, info, txnid, TXN_IGNORE, lsnp)); make_call = 1; if (ret == TXN_OK && (ret = __db_txnlist_update(dbenv, info, txnid, rectype == DB_txn_xa_regop ? TXN_PREPARE : TXN_ABORT, NULL)) != 0) return (ret); } } break; case DB_TXN_FORWARD_ROLL: /* * In the forward pass, if we haven't seen the transaction, * do nothing, else recovery it. * * We need to always redo DB_db_noop records, so that we * properly handle any commits after the file was closed. */ if (rectype == DB_log_register || rectype == DB_txn_ckp || rectype == DB_db_noop || __db_txnlist_find(dbenv, info, txnid) == TXN_COMMIT) make_call = 1; break; case DB_TXN_APPLY: make_call = 1; break; default: return (__db_unknown_flag(dbenv, "__db_dispatch", redo)); } if (make_call) { if (rectype >= DB_user_BEGIN && dbenv->tx_recover != NULL) return (dbenv->tx_recover(dbenv, db, lsnp, redo)); else if (dtab != NULL) return (dtab[rectype](dbenv, db, lsnp, redo, info)); else return (dbenv->dtab[rectype](dbenv, db, lsnp, redo, info)); } return (0); } /* * __db_add_recovery -- * * PUBLIC: int __db_add_recovery __P((DB_ENV *, * PUBLIC: int (***)(DB_ENV *, DBT *, DB_LSN *, db_recops, void *), size_t *, * PUBLIC: int (*)(DB_ENV *, DBT *, DB_LSN *, db_recops, void *), u_int32_t)); */ int __db_add_recovery(dbenv, dtab, dtabsize, func, ndx) DB_ENV *dbenv; int (***dtab) __P((DB_ENV *, DBT *, DB_LSN *, db_recops, void *)); size_t *dtabsize; int (*func) __P((DB_ENV *, DBT *, DB_LSN *, db_recops, void *)); u_int32_t ndx; { u_int32_t i, nsize; int ret; /* Check if we have to grow the table. */ if (ndx >= *dtabsize) { nsize = ndx + 40; if ((ret = __os_realloc(dbenv, nsize * sizeof((*dtab)[0]), dtab)) != 0) return (ret); for (i = *dtabsize; i < nsize; ++i) (*dtab)[i] = NULL; *dtabsize = nsize; } (*dtab)[ndx] = func; return (0); } /* * __db_txnlist_init -- * Initialize transaction linked list. * * PUBLIC: int __db_txnlist_init __P((DB_ENV *, u_int32_t, u_int32_t, void *)); */ int __db_txnlist_init(dbenv, low_txn, hi_txn, retp) DB_ENV *dbenv; u_int32_t low_txn, hi_txn; void *retp; { DB_TXNHEAD *headp; u_int32_t tmp; int ret, size; /* * Size a hash table. * If low is zero then we are being called during rollback * and we need only one slot. * Hi maybe lower than low if we have recycled txnid's. * The numbers here are guesses about txn density, we can afford * to look at a few entries in each slot. */ if (low_txn == 0) size = 1; else { if (hi_txn < low_txn) { tmp = hi_txn; hi_txn = low_txn; low_txn = tmp; } tmp = hi_txn - low_txn; /* See if we wrapped around. */ if (tmp > (TXN_INVALID - TXN_MINIMUM) / 2) tmp = (low_txn - TXN_MINIMUM) + (TXN_INVALID - hi_txn); size = tmp / 5; if (size < 100) size = 100; } if ((ret = __os_malloc(dbenv, sizeof(DB_TXNHEAD) + size * sizeof(headp->head), &headp)) != 0) return (ret); memset(headp, 0, sizeof(DB_TXNHEAD) + size * sizeof(headp->head)); headp->maxid = hi_txn; headp->generation = 1; headp->nslots = size; ZERO_LSN(headp->maxlsn); ZERO_LSN(headp->ckplsn); *(void **)retp = headp; return (0); } /* * __db_txnlist_add -- * Add an element to our transaction linked list. * * PUBLIC: int __db_txnlist_add __P((DB_ENV *, * PUBLIC: void *, u_int32_t, int32_t, DB_LSN *)); */ int __db_txnlist_add(dbenv, listp, txnid, status, lsn) DB_ENV *dbenv; void *listp; u_int32_t txnid; int32_t status; DB_LSN *lsn; { DB_TXNHEAD *hp; DB_TXNLIST *elp; int ret; if ((ret = __os_malloc(dbenv, sizeof(DB_TXNLIST), &elp)) != 0) return (ret); hp = (DB_TXNHEAD *)listp; LIST_INSERT_HEAD(&hp->head[DB_TXNLIST_MASK(hp, txnid)], elp, links); elp->type = TXNLIST_TXNID; elp->u.t.txnid = txnid; elp->u.t.status = status; elp->u.t.generation = hp->generation; if (txnid > hp->maxid) hp->maxid = txnid; if (lsn != NULL && IS_ZERO_LSN(hp->maxlsn) && status == TXN_COMMIT) hp->maxlsn = *lsn; DB_ASSERT(lsn == NULL || status != TXN_COMMIT || log_compare(&hp->maxlsn, lsn) >= 0); return (0); } /* * __db_txnlist_remove -- * Remove an element from our transaction linked list. * * PUBLIC: int __db_txnlist_remove __P((DB_ENV *, void *, u_int32_t)); */ int __db_txnlist_remove(dbenv, listp, txnid) DB_ENV *dbenv; void *listp; u_int32_t txnid; { DB_TXNLIST *entry; return (__db_txnlist_find_internal(dbenv, listp, TXNLIST_TXNID, txnid, NULL, &entry, 1) == TXN_NOTFOUND ? TXN_NOTFOUND : TXN_OK); } /* * __db_txnlist_ckp -- * Used to record the maximum checkpoint that will be retained * after recovery. Typically this is simply the max checkpoint, but * if we are doing client replication recovery or timestamp-based * recovery, we are going to virtually truncate the log and we need * to retain the last checkpoint before the truncation point. * * PUBLIC: void __db_txnlist_ckp __P((DB_ENV *, void *, DB_LSN *)); */ void __db_txnlist_ckp(dbenv, listp, ckp_lsn) DB_ENV *dbenv; void *listp; DB_LSN *ckp_lsn; { DB_TXNHEAD *hp; COMPQUIET(dbenv, NULL); hp = (DB_TXNHEAD *)listp; if (IS_ZERO_LSN(hp->ckplsn) && !IS_ZERO_LSN(hp->maxlsn) && log_compare(&hp->maxlsn, ckp_lsn) >= 0) hp->ckplsn = *ckp_lsn; } /* __db_txnlist_close -- * * Call this when we close a file. It allows us to reconcile whether * we have done any operations on this file with whether the file appears * to have been deleted. If you never do any operations on a file, then * we assume it's OK to appear deleted. * * PUBLIC: int __db_txnlist_close __P((void *, int32_t, u_int32_t)); */ int __db_txnlist_close(listp, lid, count) void *listp; int32_t lid; u_int32_t count; { DB_TXNHEAD *hp; DB_TXNLIST *p; int i; if (listp == NULL) return (0); hp = (DB_TXNHEAD *)listp; for (i = 0; i < hp->nslots; i++) for (p = LIST_FIRST(&hp->head[i]); p != NULL; p = LIST_NEXT(p, links)) { if (p->type == TXNLIST_DELETE) if (lid == p->u.d.fileid && !F_ISSET(&p->u.d, TXNLIST_FLAG_CLOSED)) { p->u.d.count += count; return (0); } } return (0); } /* * __db_txnlist_delete -- * * Record that a file was missing or deleted. If the deleted * flag is set, then we've encountered a delete of a file, else we've * just encountered a file that is missing. The lid is the log fileid * and is only meaningful if deleted is not equal to 0. * * PUBLIC: int __db_txnlist_delete __P((DB_ENV *, * PUBLIC: void *, char *, u_int32_t, int)); */ int __db_txnlist_delete(dbenv, listp, name, lid, deleted) DB_ENV *dbenv; void *listp; char *name; u_int32_t lid; int deleted; { DB_TXNHEAD *hp; DB_TXNLIST *p; int ret; hp = (DB_TXNHEAD *)listp; for (p = LIST_FIRST(&hp->head[DB_TXNLIST_MASK(hp, *name)]); p != NULL; p = LIST_NEXT(p, links)) { if (p->type == TXNLIST_DELETE) if (strcmp(name, p->u.d.fname) == 0) { if (deleted) F_SET(&p->u.d, TXNLIST_FLAG_DELETED); else F_CLR(&p->u.d, TXNLIST_FLAG_CLOSED); return (0); } } /* Need to add it. */ if ((ret = __os_malloc(dbenv, sizeof(DB_TXNLIST), &p)) != 0) return (ret); LIST_INSERT_HEAD(&hp->head[DB_TXNLIST_MASK(hp, *name)], p, links); p->type = TXNLIST_DELETE; p->u.d.flags = 0; if (deleted) F_SET(&p->u.d, TXNLIST_FLAG_DELETED); p->u.d.fileid = lid; p->u.d.count = 0; ret = __os_strdup(dbenv, name, &p->u.d.fname); return (ret); } /* * __db_txnlist_end -- * Discard transaction linked list. Print out any error messages * for deleted files. * * PUBLIC: void __db_txnlist_end __P((DB_ENV *, void *)); */ void __db_txnlist_end(dbenv, listp) DB_ENV *dbenv; void *listp; { DB_TXNHEAD *hp; DB_TXNLIST *p; DB_LOG *lp; int i; if ((hp = (DB_TXNHEAD *)listp) == NULL) return; lp = (DB_LOG *)dbenv->lg_handle; for (i = 0; i < hp->nslots; i++) while (hp != NULL && (p = LIST_FIRST(&hp->head[i])) != NULL) { LIST_REMOVE(p, links); switch (p->type) { case TXNLIST_DELETE: /* * If we have a file that is not deleted and has * some operations, we flag the warning. Since * the file could still be open, we need to * check the actual log table as well. */ if ((!F_ISSET(&p->u.d, TXNLIST_FLAG_DELETED) && p->u.d.count != 0) || (!F_ISSET(&p->u.d, TXNLIST_FLAG_CLOSED) && p->u.d.fileid != (int32_t)TXNLIST_INVALID_ID && p->u.d.fileid < lp->dbentry_cnt && lp->dbentry[p->u.d.fileid].count != 0)) __db_err(dbenv, "warning: %s: %s", p->u.d.fname, db_strerror(ENOENT)); __os_freestr(dbenv, p->u.d.fname); break; case TXNLIST_LSN: __os_free(dbenv, p->u.l.lsn_array, p->u.l.maxn * sizeof(DB_LSN)); break; default: /* * Possibly an incomplete DB_TXNLIST; just * free it. */ break; } __os_free(dbenv, p, sizeof(DB_TXNLIST)); } __os_free(dbenv, listp, sizeof(DB_TXNHEAD) + sizeof(hp->head) * hp->nslots); } /* * __db_txnlist_find -- * Checks to see if a txnid with the current generation is in the * txnid list. This returns TXN_NOTFOUND if the item isn't in the * list otherwise it returns (like __db_txnlist_find_internal) * the status of the transaction. A txnid of 0 means the record * was generated while not in a transaction. * * PUBLIC: int __db_txnlist_find __P((DB_ENV *, void *, u_int32_t)); */ int __db_txnlist_find(dbenv, listp, txnid) DB_ENV *dbenv; void *listp; u_int32_t txnid; { DB_TXNLIST *entry; if (txnid == 0) return (TXN_NOTFOUND); return (__db_txnlist_find_internal(dbenv, listp, TXNLIST_TXNID, txnid, NULL, &entry, 0)); } /* * __db_txnlist_update -- * Change the status of an existing transaction entry. * Returns TXN_NOTFOUND if no such entry exists. * * PUBLIC: int __db_txnlist_update __P((DB_ENV *, * PUBLIC: void *, u_int32_t, u_int32_t, DB_LSN *)); */ int __db_txnlist_update(dbenv, listp, txnid, status, lsn) DB_ENV *dbenv; void *listp; u_int32_t txnid; u_int32_t status; DB_LSN *lsn; { DB_TXNHEAD *hp; DB_TXNLIST *elp; int ret; if (txnid == 0) return (TXN_NOTFOUND); hp = (DB_TXNHEAD *)listp; ret = __db_txnlist_find_internal(dbenv, listp, TXNLIST_TXNID, txnid, NULL, &elp, 0); if (ret == TXN_NOTFOUND) return (ret); elp->u.t.status = status; if (lsn != NULL && IS_ZERO_LSN(hp->maxlsn) && status == TXN_COMMIT) hp->maxlsn = *lsn; return (ret); } /* * __db_txnlist_find_internal -- * Find an entry on the transaction list. If the entry is not there or * the list pointer is not initialized we return TXN_NOTFOUND. If the * item is found, we return the status. Currently we always call this * with an initialized list pointer but checking for NULL keeps it general. */ static int __db_txnlist_find_internal(dbenv, listp, type, txnid, uid, txnlistp, delete) DB_ENV *dbenv; void *listp; db_txnlist_type type; u_int32_t txnid; u_int8_t uid[DB_FILE_ID_LEN]; DB_TXNLIST **txnlistp; int delete; { DB_TXNHEAD *hp; DB_TXNLIST *p; u_int32_t hash; struct __db_headlink *head; int ret; if ((hp = (DB_TXNHEAD *)listp) == NULL) return (TXN_NOTFOUND); switch (type) { case TXNLIST_TXNID: hash = txnid; break; case TXNLIST_PGNO: memcpy(&hash, uid, sizeof(hash)); break; default: DB_ASSERT(0); ret = EINVAL; } head = &hp->head[DB_TXNLIST_MASK(hp, hash)]; for (p = LIST_FIRST(head); p != NULL; p = LIST_NEXT(p, links)) { if (p->type != type) continue; switch (type) { case TXNLIST_TXNID: if (p->u.t.txnid != txnid || hp->generation != p->u.t.generation) continue; ret = p->u.t.status; break; case TXNLIST_PGNO: if (memcmp(uid, p->u.p.uid, DB_FILE_ID_LEN) != 0) continue; ret = 0; break; default: DB_ASSERT(0); ret = EINVAL; } if (delete == 1) { LIST_REMOVE(p, links); __os_free(dbenv, p, sizeof(DB_TXNLIST)); } else if (p != LIST_FIRST(head)) { /* Move it to head of list. */ LIST_REMOVE(p, links); LIST_INSERT_HEAD(head, p, links); } *txnlistp = p; return (ret); } return (TXN_NOTFOUND); } /* * __db_txnlist_gen -- * Change the current generation number. * * PUBLIC: void __db_txnlist_gen __P((void *, int)); */ void __db_txnlist_gen(listp, incr) void *listp; int incr; { DB_TXNHEAD *hp; /* * During recovery generation numbers keep track of how many "restart" * checkpoints we've seen. Restart checkpoints occur whenever we take * a checkpoint and there are no outstanding transactions. When that * happens, we can reset transaction IDs back to 1. It always happens * at recovery and it prevents us from exhausting the transaction IDs * name space. */ hp = (DB_TXNHEAD *)listp; hp->generation += incr; } #define TXN_BUBBLE(AP, MAX) { \ int __j; \ DB_LSN __tmp; \ \ for (__j = 0; __j < MAX - 1; __j++) \ if (log_compare(&AP[__j], &AP[__j + 1]) < 0) { \ __tmp = AP[__j]; \ AP[__j] = AP[__j + 1]; \ AP[__j + 1] = __tmp; \ } \ } /* * __db_txnlist_lsnadd -- * Add to or re-sort the transaction list lsn entry. Note that since this * is used during an abort, the __txn_undo code calls into the "recovery" * subsystem explicitly, and there is only a single TXNLIST_LSN entry on * the list. * * PUBLIC: int __db_txnlist_lsnadd __P((DB_ENV *, void *, DB_LSN *, u_int32_t)); */ int __db_txnlist_lsnadd(dbenv, listp, lsnp, flags) DB_ENV *dbenv; void *listp; DB_LSN *lsnp; u_int32_t flags; { DB_TXNHEAD *hp; DB_TXNLIST *elp; int i, ret; hp = (DB_TXNHEAD *)listp; for (elp = LIST_FIRST(&hp->head[0]); elp != NULL; elp = LIST_NEXT(elp, links)) if (elp->type == TXNLIST_LSN) break; if (elp == NULL) return (DB_SURPRISE_KID); if (LF_ISSET(TXNLIST_NEW)) { if (elp->u.l.ntxns >= elp->u.l.maxn) { if ((ret = __os_realloc(dbenv, 2 * elp->u.l.maxn * sizeof(DB_LSN), &elp->u.l.lsn_array)) != 0) return (ret); elp->u.l.maxn *= 2; } elp->u.l.lsn_array[elp->u.l.ntxns++] = *lsnp; } else /* Simply replace the 0th element. */ elp->u.l.lsn_array[0] = *lsnp; /* * If we just added a new entry and there may be NULL entries, so we * have to do a complete bubble sort, not just trickle a changed entry * around. */ for (i = 0; i < (!LF_ISSET(TXNLIST_NEW) ? 1 : elp->u.l.ntxns); i++) TXN_BUBBLE(elp->u.l.lsn_array, elp->u.l.ntxns); *lsnp = elp->u.l.lsn_array[0]; return (0); } /* * __db_txnlist_lsninit -- * Initialize a transaction list with an lsn array entry. * * PUBLIC: int __db_txnlist_lsninit __P((DB_ENV *, DB_TXNHEAD *, DB_LSN *)); */ int __db_txnlist_lsninit(dbenv, hp, lsnp) DB_ENV *dbenv; DB_TXNHEAD *hp; DB_LSN *lsnp; { DB_TXNLIST *elp; int ret; elp = NULL; if ((ret = __os_malloc(dbenv, sizeof(DB_TXNLIST), &elp)) != 0) goto err; LIST_INSERT_HEAD(&hp->head[0], elp, links); if ((ret = __os_malloc(dbenv, 12 * sizeof(DB_LSN), &elp->u.l.lsn_array)) != 0) goto err; elp->type = TXNLIST_LSN; elp->u.l.maxn = 12; elp->u.l.ntxns = 1; elp->u.l.lsn_array[0] = *lsnp; return (0); err: __db_txnlist_end(dbenv, hp); return (ret); } /* * __db_add_limbo -- add pages to the limbo list. * Get the file information and call pgnoadd for each page. * * PUBLIC: int __db_add_limbo __P((DB_ENV *, * PUBLIC: void *, int32_t, db_pgno_t, int32_t)); */ int __db_add_limbo(dbenv, info, fileid, pgno, count) DB_ENV *dbenv; void *info; int32_t fileid; db_pgno_t pgno; int32_t count; { DB_LOG *dblp; FNAME *fnp; int ret; dblp = dbenv->lg_handle; if ((ret = __log_lid_to_fname(dblp, fileid, &fnp)) != 0) return (ret); do { if ((ret = __db_txnlist_pgnoadd(dbenv, info, fileid, fnp->ufid, R_ADDR(&dblp->reginfo, fnp->name_off), pgno)) != 0) return (ret); pgno++; } while (--count != 0); return (0); } /* * __db_do_the_limbo -- move pages from limbo to free. * * If we are in recovery we add things to the free list without logging * because we want to incrementaly apply logs that may be generated on * another copy of this environment. Otherwise we just call __db_free to * put the pages on the free list and log the activity. * * PUBLIC: int __db_do_the_limbo __P((DB_ENV *, DB_TXNHEAD *)); */ int __db_do_the_limbo(dbenv, hp) DB_ENV *dbenv; DB_TXNHEAD *hp; { DB *dbp; DBC *dbc; DBMETA *meta; DB_MPOOLFILE *mpf; DB_TXN *txn; DB_TXNLIST *elp; PAGE *pagep; db_pgno_t last_pgno, pgno; int h, i, in_recover, put_page, ret, t_ret; dbp = NULL; dbc = NULL; txn = NULL; ret = 0; /* Are we in recovery? */ in_recover = F_ISSET((DB_LOG *)dbenv->lg_handle, DBLOG_RECOVER) ? 1 : 0; for (h = 0; h < hp->nslots; h++) for (elp = LIST_FIRST(&hp->head[h]); elp != NULL; elp = LIST_NEXT(elp, links)) { if (elp->type != TXNLIST_PGNO) continue; if (in_recover != 0) { if ((ret = db_create(&dbp, dbenv, 0)) != 0) goto err; /* * It is ok if the file is nolonger there. */ dbp->type = DB_UNKNOWN; ret = __db_dbopen(dbp, elp->u.p.fname, DB_ODDFILESIZE, __db_omode("rw----"), 0); } else { /* * If we are in transaction undo, then we know * the fileid is still correct. */ if ((ret = __db_fileid_to_db(dbenv, &dbp, elp->u.p.fileid, 0)) != 0 && ret != DB_DELETED) goto err; /* File is being destroyed. */ if (F_ISSET(dbp, DB_AM_DISCARD)) ret = DB_DELETED; } /* * Verify that we are opening the same file that we were * referring to when we wrote this log record. */ if (ret == 0 && memcmp(elp->u.p.uid, dbp->fileid, DB_FILE_ID_LEN) == 0) { mpf = dbp->mpf; last_pgno = PGNO_INVALID; recover: if (in_recover != 0) { pgno = PGNO_BASE_MD; if ((ret = mpf->get( mpf, &pgno, 0, (PAGE **)&meta)) != 0) goto err; last_pgno = meta->free; /* * Check to see if the head of the free * list is any of the pages we are about * to link in. We could have crashed * after linking them in and before writing * a checkpoint. * It may not be the last one since * any page may get reallocated before here. */ if (last_pgno != PGNO_INVALID) for (i = 0; i < elp->u.p.nentries; i++) if (last_pgno == elp->u.p.pgno_array[i]) goto done_it; } for (i = 0; i < elp->u.p.nentries; i++) { pgno = elp->u.p.pgno_array[i]; if ((ret = mpf->get(mpf, &pgno, DB_MPOOL_CREATE, &pagep)) != 0) goto err; put_page = 1; if (IS_ZERO_LSN(LSN(pagep))) { P_INIT(pagep, dbp->pgsize, pgno, PGNO_INVALID, last_pgno, 0, P_INVALID); if (in_recover != 0) { LSN(pagep) = LSN(meta); last_pgno = pgno; } else { /* * Starting the transaction * is postponed until we know * we have something to do. */ if (txn == NULL && (ret = __txn_compensate_begin( dbenv, &txn)) != 0) goto err; if (dbc == NULL && (ret = dbp->cursor(dbp, txn, &dbc, 0)) != 0) goto err; /* Turn off locking. */ F_SET(dbc, DBC_COMPENSATE); /* __db_free puts the page. */ ret = __db_free(dbc, pagep); if (ret == EIO) { /* * Log is probably * out of space. */ in_recover = 2; (void)dbc->c_close(dbc); dbc = NULL; (void)txn->abort(txn); txn = NULL; ret = 0; goto recover; } put_page = 0; } } if (put_page == 1 && (ret = mpf->put(mpf, pagep, DB_MPOOL_DIRTY)) != 0) goto err; } if (in_recover != 0) { if (last_pgno == meta->free) { done_it: if ((ret = mpf->put(mpf, meta, 0)) != 0) goto err; } else { /* * Flush the new free list then * update the metapage. This is * unlogged so we cannot have the * metapage pointing at pages that * are not on disk. */ if (in_recover == 2) __db_err(dbenv, "Error in abort of an allocation, flushing."); dbp->sync(dbp, 0); meta->free = last_pgno; if ((ret = mpf->put( mpf, meta, DB_MPOOL_DIRTY)) != 0) goto err; } } if (dbc != NULL && (ret = dbc->c_close(dbc)) != 0) goto err; dbc = NULL; } if (in_recover == 1 && (t_ret = dbp->close(dbp, 0)) != 0 && ret == 0) ret = t_ret; dbp = NULL; __os_free(dbenv, elp->u.p.fname, 0); __os_free(dbenv, elp->u.p.pgno_array, 0); if (ret == ENOENT) ret = 0; else if (ret != 0) goto err; } if (txn != NULL) { ret = txn->commit(txn, DB_TXN_NOSYNC); txn = NULL; } err: if (ret == DB_DELETED) ret = 0; if (ret != 0) { __db_err(dbenv, "Fatal error in abort of an allocation."); ret = __db_panic(dbenv, ret); } if (dbc != NULL) (void)dbc->c_close(dbc); if (in_recover == 1 && dbp != NULL) (void)dbp->close(dbp, 0); if (txn != NULL) (void)txn->abort(txn); return (ret); } #define DB_TXNLIST_MAX_PGNO 8 /* A nice even number. */ /* * __db_txnlist_pgnoadd -- * Find the txnlist entry for a file and add this pgno, or add the list * entry for the file and then add the pgno. */ static int __db_txnlist_pgnoadd(dbenv, hp, fileid, uid, fname, pgno) DB_ENV *dbenv; DB_TXNHEAD *hp; int32_t fileid; u_int8_t uid[DB_FILE_ID_LEN]; char *fname; db_pgno_t pgno; { DB_TXNLIST *elp; u_int32_t hash; int len, ret; elp = NULL; if (__db_txnlist_find_internal(dbenv, hp, TXNLIST_PGNO, 0, uid, &elp, 0) != 0) { if ((ret = __os_malloc(dbenv, sizeof(DB_TXNLIST), &elp)) != 0) goto err; memcpy(&hash, uid, sizeof(hash)); LIST_INSERT_HEAD( &hp->head[DB_TXNLIST_MASK(hp, hash)], elp, links); elp->u.p.fileid = fileid; memcpy(elp->u.p.uid, uid, DB_FILE_ID_LEN); len = strlen(fname) + 1; if ((ret = __os_malloc(dbenv, len, &elp->u.p.fname)) != 0) goto err; memcpy(elp->u.p.fname, fname, len); elp->u.p.maxentry = 0; elp->type = TXNLIST_PGNO; if ((ret = __os_malloc(dbenv, 8 * sizeof(db_pgno_t), &elp->u.p.pgno_array)) != 0) goto err; elp->u.p.maxentry = DB_TXNLIST_MAX_PGNO; elp->u.p.nentries = 0; } else if (elp->u.p.nentries == elp->u.p.maxentry) { elp->u.p.maxentry <<= 1; if ((ret = __os_realloc(dbenv, elp->u.p.maxentry * sizeof(db_pgno_t), &elp->u.p.pgno_array)) != 0) goto err; } elp->u.p.pgno_array[elp->u.p.nentries++] = pgno; return (0); err: __db_txnlist_end(dbenv, hp); return (ret); } #ifdef DEBUG /* * __db_txnlist_print -- * Print out the transaction list. * * PUBLIC: void __db_txnlist_print __P((void *)); */ void __db_txnlist_print(listp) void *listp; { DB_TXNHEAD *hp; DB_TXNLIST *p; int i; hp = (DB_TXNHEAD *)listp; printf("Maxid: %lu Generation: %lu\n", (u_long)hp->maxid, (u_long)hp->generation); for (i = 0; i < hp->nslots; i++) for (p = LIST_FIRST(&hp->head[i]); p != NULL; p = LIST_NEXT(p, links)) { switch (p->type) { case TXNLIST_TXNID: printf("TXNID: %lu(%lu)\n", (u_long)p->u.t.txnid, (u_long)p->u.t.generation); break; case TXNLIST_DELETE: printf("FILE: %s id=%d ops=%d %s %s\n", p->u.d.fname, (int)p->u.d.fileid, p->u.d.count, F_ISSET(&p->u.d, TXNLIST_FLAG_DELETED) ? "(deleted)" : "(missing)", F_ISSET(&p->u.d, TXNLIST_FLAG_CLOSED) ? "(closed)" : "(open)"); break; default: printf("Unrecognized type: %d\n", p->type); break; } } } #endif