summaryrefslogtreecommitdiff
path: root/bdb/rep/rep_method.c
diff options
context:
space:
mode:
authorunknown <ram@mysql.r18.ru>2002-10-30 15:57:05 +0400
committerunknown <ram@mysql.r18.ru>2002-10-30 15:57:05 +0400
commit155e78f014de1a2e259ae5119f4621fbb210a784 (patch)
tree6881a3cca88bea0bb9eeffd5aae34be437152786 /bdb/rep/rep_method.c
parentb8798d25ab71436bf690ee8ae48285a655c5487e (diff)
downloadmariadb-git-155e78f014de1a2e259ae5119f4621fbb210a784.tar.gz
BDB 4.1.24
BitKeeper/deleted/.del-ex_access.wpj~3df6ae8c99bf7c5f: Delete: bdb/build_vxworks/ex_access/ex_access.wpj BitKeeper/deleted/.del-ex_btrec.wpj~a7622f1c6f432dc6: Delete: bdb/build_vxworks/ex_btrec/ex_btrec.wpj BitKeeper/deleted/.del-ex_dbclient.wpj~7345440f3b204cdd: Delete: bdb/build_vxworks/ex_dbclient/ex_dbclient.wpj BitKeeper/deleted/.del-ex_env.wpj~fbe1ab10b04e8b74: Delete: bdb/build_vxworks/ex_env/ex_env.wpj BitKeeper/deleted/.del-ex_mpool.wpj~4479cfd5c45f327d: Delete: bdb/build_vxworks/ex_mpool/ex_mpool.wpj BitKeeper/deleted/.del-ex_tpcb.wpj~f78093006e14bf41: Delete: bdb/build_vxworks/ex_tpcb/ex_tpcb.wpj BitKeeper/deleted/.del-db_buildall.dsp~bd749ff6da11682: Delete: bdb/build_win32/db_buildall.dsp BitKeeper/deleted/.del-cxx_app.cpp~ad8df8e0791011ed: Delete: bdb/cxx/cxx_app.cpp BitKeeper/deleted/.del-cxx_log.cpp~a50ff3118fe06952: Delete: bdb/cxx/cxx_log.cpp BitKeeper/deleted/.del-cxx_table.cpp~ecd751e79b055556: Delete: bdb/cxx/cxx_table.cpp BitKeeper/deleted/.del-namemap.txt~796a3acd3885d8fd: Delete: bdb/cxx/namemap.txt BitKeeper/deleted/.del-Design.fileop~3ca4da68f1727373: Delete: bdb/db/Design.fileop BitKeeper/deleted/.del-db185_int.h~61bee3736e7959ef: Delete: bdb/db185/db185_int.h BitKeeper/deleted/.del-acconfig.h~411e8854d67ad8b5: Delete: bdb/dist/acconfig.h BitKeeper/deleted/.del-mutex.m4~a13383cde18a64e1: Delete: bdb/dist/aclocal/mutex.m4 BitKeeper/deleted/.del-options.m4~b9d0ca637213750a: Delete: bdb/dist/aclocal/options.m4 BitKeeper/deleted/.del-programs.m4~3ce7890b47732b30: Delete: bdb/dist/aclocal/programs.m4 BitKeeper/deleted/.del-tcl.m4~f944e2db93c3b6db: Delete: bdb/dist/aclocal/tcl.m4 BitKeeper/deleted/.del-types.m4~59cae158c9a32cff: Delete: bdb/dist/aclocal/types.m4 BitKeeper/deleted/.del-script~d38f6d3a4f159cb4: Delete: bdb/dist/build/script BitKeeper/deleted/.del-configure.in~ac795a92c8fe049c: Delete: bdb/dist/configure.in BitKeeper/deleted/.del-ltconfig~66bbd007d8024af: Delete: bdb/dist/ltconfig BitKeeper/deleted/.del-rec_ctemp~a28554362534f00a: Delete: bdb/dist/rec_ctemp BitKeeper/deleted/.del-s_tcl~2ffe4326459fcd9f: Delete: bdb/dist/s_tcl BitKeeper/deleted/.del-.IGNORE_ME~d8148b08fa7d5d15: Delete: bdb/dist/template/.IGNORE_ME BitKeeper/deleted/.del-btree.h~179f2aefec1753d: Delete: bdb/include/btree.h BitKeeper/deleted/.del-cxx_int.h~6b649c04766508f8: Delete: bdb/include/cxx_int.h BitKeeper/deleted/.del-db.src~6b433ae615b16a8d: Delete: bdb/include/db.src BitKeeper/deleted/.del-db_185.h~ad8b373d9391d35c: Delete: bdb/include/db_185.h BitKeeper/deleted/.del-db_am.h~a714912b6b75932f: Delete: bdb/include/db_am.h BitKeeper/deleted/.del-db_cxx.h~fcafadf45f5d19e9: Delete: bdb/include/db_cxx.h BitKeeper/deleted/.del-db_dispatch.h~6844f20f7eb46904: Delete: bdb/include/db_dispatch.h BitKeeper/deleted/.del-db_int.src~419a3f48b6a01da7: Delete: bdb/include/db_int.src BitKeeper/deleted/.del-db_join.h~76f9747a42c3399a: Delete: bdb/include/db_join.h BitKeeper/deleted/.del-db_page.h~e302ca3a4db3abdc: Delete: bdb/include/db_page.h BitKeeper/deleted/.del-db_server_int.h~e1d20b6ba3bca1ab: Delete: bdb/include/db_server_int.h BitKeeper/deleted/.del-db_shash.h~5fbf2d696fac90f3: Delete: bdb/include/db_shash.h BitKeeper/deleted/.del-db_swap.h~1e60887550864a59: Delete: bdb/include/db_swap.h BitKeeper/deleted/.del-db_upgrade.h~c644eee73701fc8d: Delete: bdb/include/db_upgrade.h BitKeeper/deleted/.del-db_verify.h~b8d6c297c61f342e: Delete: bdb/include/db_verify.h BitKeeper/deleted/.del-debug.h~dc2b4f2cf27ccebc: Delete: bdb/include/debug.h BitKeeper/deleted/.del-hash.h~2aaa548b28882dfb: Delete: bdb/include/hash.h BitKeeper/deleted/.del-lock.h~a761c1b7de57b77f: Delete: bdb/include/lock.h BitKeeper/deleted/.del-log.h~ff20184238e35e4d: Delete: bdb/include/log.h BitKeeper/deleted/.del-mp.h~7e317597622f3411: Delete: bdb/include/mp.h BitKeeper/deleted/.del-mutex.h~d3ae7a2977a68137: Delete: bdb/include/mutex.h BitKeeper/deleted/.del-os.h~91867cc8757cd0e3: Delete: bdb/include/os.h BitKeeper/deleted/.del-os_jump.h~e1b939fa5151d4be: Delete: bdb/include/os_jump.h BitKeeper/deleted/.del-qam.h~6fad0c1b5723d597: Delete: bdb/include/qam.h BitKeeper/deleted/.del-queue.h~4c72c0826c123d5: Delete: bdb/include/queue.h BitKeeper/deleted/.del-region.h~513fe04d977ca0fc: Delete: bdb/include/region.h BitKeeper/deleted/.del-shqueue.h~525fc3e6c2025c36: Delete: bdb/include/shqueue.h BitKeeper/deleted/.del-tcl_db.h~c536fd61a844f23f: Delete: bdb/include/tcl_db.h BitKeeper/deleted/.del-txn.h~c8d94b221ec147e4: Delete: bdb/include/txn.h BitKeeper/deleted/.del-xa.h~ecc466493aae9d9a: Delete: bdb/include/xa.h BitKeeper/deleted/.del-DbRecoveryInit.java~756b52601a0b9023: Delete: bdb/java/src/com/sleepycat/db/DbRecoveryInit.java BitKeeper/deleted/.del-DbTxnRecover.java~74607cba7ab89d6d: Delete: bdb/java/src/com/sleepycat/db/DbTxnRecover.java BitKeeper/deleted/.del-lock_conflict.c~fc5e0f14cf597a2b: Delete: bdb/lock/lock_conflict.c BitKeeper/deleted/.del-log.src~53ac9e7b5cb023f2: Delete: bdb/log/log.src BitKeeper/deleted/.del-log_findckp.c~24287f008916e81f: Delete: bdb/log/log_findckp.c BitKeeper/deleted/.del-log_rec.c~d51711f2cac09297: Delete: bdb/log/log_rec.c BitKeeper/deleted/.del-log_register.c~b40bb4efac75ca15: Delete: bdb/log/log_register.c BitKeeper/deleted/.del-Design~b3d0f179f2767b: Delete: bdb/mp/Design BitKeeper/deleted/.del-os_finit.c~95dbefc6fe79b26c: Delete: bdb/os/os_finit.c BitKeeper/deleted/.del-os_abs.c~df95d1e7db81924: Delete: bdb/os_vxworks/os_abs.c BitKeeper/deleted/.del-os_finit.c~803b484bdb9d0122: Delete: bdb/os_vxworks/os_finit.c BitKeeper/deleted/.del-os_map.c~3a6d7926398b76d3: Delete: bdb/os_vxworks/os_map.c BitKeeper/deleted/.del-os_finit.c~19a227c6d3c78ad: Delete: bdb/os_win32/os_finit.c BitKeeper/deleted/.del-log-corruption.patch~1cf2ecc7c6408d5d: Delete: bdb/patches/log-corruption.patch BitKeeper/deleted/.del-Btree.pm~af6d0c5eaed4a98e: Delete: bdb/perl.BerkeleyDB/BerkeleyDB/Btree.pm BitKeeper/deleted/.del-BerkeleyDB.pm~7244036d4482643: Delete: bdb/perl.BerkeleyDB/BerkeleyDB.pm BitKeeper/deleted/.del-BerkeleyDB.pod~e7b18fd6132448e3: Delete: bdb/perl.BerkeleyDB/BerkeleyDB.pod BitKeeper/deleted/.del-Hash.pm~10292a26c06a5c95: Delete: bdb/perl.BerkeleyDB/BerkeleyDB/Hash.pm BitKeeper/deleted/.del-BerkeleyDB.pod.P~79f76a1495eda203: Delete: bdb/perl.BerkeleyDB/BerkeleyDB.pod.P BitKeeper/deleted/.del-BerkeleyDB.xs~80c99afbd98e392c: Delete: bdb/perl.BerkeleyDB/BerkeleyDB.xs BitKeeper/deleted/.del-Changes~729c1891efa60de9: Delete: bdb/perl.BerkeleyDB/Changes BitKeeper/deleted/.del-MANIFEST~63a1e34aecf157a0: Delete: bdb/perl.BerkeleyDB/MANIFEST BitKeeper/deleted/.del-Makefile.PL~c68797707d8df87a: Delete: bdb/perl.BerkeleyDB/Makefile.PL BitKeeper/deleted/.del-README~5f2f579b1a241407: Delete: bdb/perl.BerkeleyDB/README BitKeeper/deleted/.del-Todo~dca3c66c193adda9: Delete: bdb/perl.BerkeleyDB/Todo BitKeeper/deleted/.del-config.in~ae81681e450e0999: Delete: bdb/perl.BerkeleyDB/config.in BitKeeper/deleted/.del-dbinfo~28ad67d83be4f68e: Delete: bdb/perl.BerkeleyDB/dbinfo BitKeeper/deleted/.del-mkconsts~543ab60669c7a04e: Delete: bdb/perl.BerkeleyDB/mkconsts BitKeeper/deleted/.del-mkpod~182c0ca54e439afb: Delete: bdb/perl.BerkeleyDB/mkpod BitKeeper/deleted/.del-5.004~e008cb5a48805543: Delete: bdb/perl.BerkeleyDB/patches/5.004 BitKeeper/deleted/.del-irix_6_5.pl~61662bb08afcdec8: Delete: bdb/perl.BerkeleyDB/hints/irix_6_5.pl BitKeeper/deleted/.del-solaris.pl~6771e7182394e152: Delete: bdb/perl.BerkeleyDB/hints/solaris.pl BitKeeper/deleted/.del-typemap~783b8f5295b05f3d: Delete: bdb/perl.BerkeleyDB/typemap BitKeeper/deleted/.del-5.004_01~6081ce2fff7b0bc: Delete: bdb/perl.BerkeleyDB/patches/5.004_01 BitKeeper/deleted/.del-5.004_02~87214eac35ad9e6: Delete: bdb/perl.BerkeleyDB/patches/5.004_02 BitKeeper/deleted/.del-5.004_03~9a672becec7cb40f: Delete: bdb/perl.BerkeleyDB/patches/5.004_03 BitKeeper/deleted/.del-5.004_04~e326cb51af09d154: Delete: bdb/perl.BerkeleyDB/patches/5.004_04 BitKeeper/deleted/.del-5.004_05~7ab457a1e41a92fe: Delete: bdb/perl.BerkeleyDB/patches/5.004_05 BitKeeper/deleted/.del-5.005~f9e2d59b5964cd4b: Delete: bdb/perl.BerkeleyDB/patches/5.005 BitKeeper/deleted/.del-5.005_01~3eb9fb7b5842ea8e: Delete: bdb/perl.BerkeleyDB/patches/5.005_01 BitKeeper/deleted/.del-5.005_02~67477ce0bef717cb: Delete: bdb/perl.BerkeleyDB/patches/5.005_02 BitKeeper/deleted/.del-5.005_03~c4c29a1fb21e290a: Delete: bdb/perl.BerkeleyDB/patches/5.005_03 BitKeeper/deleted/.del-5.6.0~e1fb9897d124ee22: Delete: bdb/perl.BerkeleyDB/patches/5.6.0 BitKeeper/deleted/.del-btree.t~e4a1a3c675ddc406: Delete: bdb/perl.BerkeleyDB/t/btree.t BitKeeper/deleted/.del-db-3.0.t~d2c60991d84558f2: Delete: bdb/perl.BerkeleyDB/t/db-3.0.t BitKeeper/deleted/.del-db-3.1.t~6ee88cd13f55e018: Delete: bdb/perl.BerkeleyDB/t/db-3.1.t BitKeeper/deleted/.del-db-3.2.t~f73b6461f98fd1cf: Delete: bdb/perl.BerkeleyDB/t/db-3.2.t BitKeeper/deleted/.del-destroy.t~cc6a2ae1980a2ecd: Delete: bdb/perl.BerkeleyDB/t/destroy.t BitKeeper/deleted/.del-env.t~a8604a4499c4bd07: Delete: bdb/perl.BerkeleyDB/t/env.t BitKeeper/deleted/.del-examples.t~2571b77c3cc75574: Delete: bdb/perl.BerkeleyDB/t/examples.t BitKeeper/deleted/.del-examples.t.T~8228bdd75ac78b88: Delete: bdb/perl.BerkeleyDB/t/examples.t.T BitKeeper/deleted/.del-examples3.t.T~66a186897a87026d: Delete: bdb/perl.BerkeleyDB/t/examples3.t.T BitKeeper/deleted/.del-examples3.t~fe3822ba2f2d7f83: Delete: bdb/perl.BerkeleyDB/t/examples3.t BitKeeper/deleted/.del-filter.t~f87b045c1b708637: Delete: bdb/perl.BerkeleyDB/t/filter.t BitKeeper/deleted/.del-hash.t~616bfb4d644de3a3: Delete: bdb/perl.BerkeleyDB/t/hash.t BitKeeper/deleted/.del-join.t~29fc39f74a83ca22: Delete: bdb/perl.BerkeleyDB/t/join.t BitKeeper/deleted/.del-mldbm.t~31f5015341eea040: Delete: bdb/perl.BerkeleyDB/t/mldbm.t BitKeeper/deleted/.del-queue.t~8f338034ce44a641: Delete: bdb/perl.BerkeleyDB/t/queue.t BitKeeper/deleted/.del-recno.t~d4ddbd3743add63e: Delete: bdb/perl.BerkeleyDB/t/recno.t BitKeeper/deleted/.del-strict.t~6885cdd2ea71ca2d: Delete: bdb/perl.BerkeleyDB/t/strict.t BitKeeper/deleted/.del-subdb.t~aab62a5d5864c603: Delete: bdb/perl.BerkeleyDB/t/subdb.t BitKeeper/deleted/.del-txn.t~65033b8558ae1216: Delete: bdb/perl.BerkeleyDB/t/txn.t BitKeeper/deleted/.del-unknown.t~f3710458682665e1: Delete: bdb/perl.BerkeleyDB/t/unknown.t BitKeeper/deleted/.del-Changes~436f74a5c414c65b: Delete: bdb/perl.DB_File/Changes BitKeeper/deleted/.del-DB_File.pm~ae0951c6c7665a82: Delete: bdb/perl.DB_File/DB_File.pm BitKeeper/deleted/.del-DB_File.xs~89e49a0b5556f1d8: Delete: bdb/perl.DB_File/DB_File.xs BitKeeper/deleted/.del-DB_File_BS~290fad5dbbb87069: Delete: bdb/perl.DB_File/DB_File_BS BitKeeper/deleted/.del-MANIFEST~90ee581572bdd4ac: Delete: bdb/perl.DB_File/MANIFEST BitKeeper/deleted/.del-Makefile.PL~ac0567bb5a377e38: Delete: bdb/perl.DB_File/Makefile.PL BitKeeper/deleted/.del-README~77e924a5a9bae6b3: Delete: bdb/perl.DB_File/README BitKeeper/deleted/.del-config.in~ab4c2792b86a810b: Delete: bdb/perl.DB_File/config.in BitKeeper/deleted/.del-dbinfo~461c43b30fab2cb: Delete: bdb/perl.DB_File/dbinfo BitKeeper/deleted/.del-dynixptx.pl~50dcddfae25d17e9: Delete: bdb/perl.DB_File/hints/dynixptx.pl BitKeeper/deleted/.del-typemap~55cffb3288a9e587: Delete: bdb/perl.DB_File/typemap BitKeeper/deleted/.del-version.c~a4df0e646f8b3975: Delete: bdb/perl.DB_File/version.c BitKeeper/deleted/.del-5.004_01~d6830d0082702af7: Delete: bdb/perl.DB_File/patches/5.004_01 BitKeeper/deleted/.del-5.004_02~78b082dc80c91031: Delete: bdb/perl.DB_File/patches/5.004_02 BitKeeper/deleted/.del-5.004~4411ec2e3c9e008b: Delete: bdb/perl.DB_File/patches/5.004 BitKeeper/deleted/.del-sco.pl~1e795fe14fe4dcfe: Delete: bdb/perl.DB_File/hints/sco.pl BitKeeper/deleted/.del-5.004_03~33f274648b160d95: Delete: bdb/perl.DB_File/patches/5.004_03 BitKeeper/deleted/.del-5.004_04~8f3d1b3cf18bb20a: Delete: bdb/perl.DB_File/patches/5.004_04 BitKeeper/deleted/.del-5.004_05~9c0f02e7331e142: Delete: bdb/perl.DB_File/patches/5.004_05 BitKeeper/deleted/.del-5.005~c2108cb2e3c8d951: Delete: bdb/perl.DB_File/patches/5.005 BitKeeper/deleted/.del-5.005_01~3b45e9673afc4cfa: Delete: bdb/perl.DB_File/patches/5.005_01 BitKeeper/deleted/.del-5.005_02~9fe5766bb02a4522: Delete: bdb/perl.DB_File/patches/5.005_02 BitKeeper/deleted/.del-5.005_03~ffa1c38c19ae72ea: Delete: bdb/perl.DB_File/patches/5.005_03 BitKeeper/deleted/.del-5.6.0~373be3a5ce47be85: Delete: bdb/perl.DB_File/patches/5.6.0 BitKeeper/deleted/.del-db-btree.t~3231595a1c241eb3: Delete: bdb/perl.DB_File/t/db-btree.t BitKeeper/deleted/.del-db-hash.t~7c4ad0c795c7fad2: Delete: bdb/perl.DB_File/t/db-hash.t BitKeeper/deleted/.del-db-recno.t~6c2d3d80b9ba4a50: Delete: bdb/perl.DB_File/t/db-recno.t BitKeeper/deleted/.del-db_server.sed~cdb00ebcd48a64e2: Delete: bdb/rpc_server/db_server.sed BitKeeper/deleted/.del-db_server_proc.c~d46c8f409c3747f4: Delete: bdb/rpc_server/db_server_proc.c BitKeeper/deleted/.del-db_server_svc.sed~3f5e59f334fa4607: Delete: bdb/rpc_server/db_server_svc.sed BitKeeper/deleted/.del-db_server_util.c~a809f3a4629acda: Delete: bdb/rpc_server/db_server_util.c BitKeeper/deleted/.del-log.tcl~ff1b41f1355b97d7: Delete: bdb/test/log.tcl BitKeeper/deleted/.del-mpool.tcl~b0df4dc1b04db26c: Delete: bdb/test/mpool.tcl BitKeeper/deleted/.del-mutex.tcl~52fd5c73a150565: Delete: bdb/test/mutex.tcl BitKeeper/deleted/.del-txn.tcl~c4ff071550b5446e: Delete: bdb/test/txn.tcl BitKeeper/deleted/.del-README~e800a12a5392010a: Delete: bdb/test/upgrade/README BitKeeper/deleted/.del-pack-2.6.6.pl~89d5076d758d3e98: Delete: bdb/test/upgrade/generate-2.X/pack-2.6.6.pl BitKeeper/deleted/.del-test-2.6.patch~4a52dc83d447547b: Delete: bdb/test/upgrade/generate-2.X/test-2.6.patch
Diffstat (limited to 'bdb/rep/rep_method.c')
-rw-r--r--bdb/rep/rep_method.c1144
1 files changed, 1144 insertions, 0 deletions
diff --git a/bdb/rep/rep_method.c b/bdb/rep/rep_method.c
new file mode 100644
index 00000000000..6773a537f4f
--- /dev/null
+++ b/bdb/rep/rep_method.c
@@ -0,0 +1,1144 @@
+/*-
+ * See the file LICENSE for redistribution information.
+ *
+ * Copyright (c) 2001-2002
+ * Sleepycat Software. All rights reserved.
+ */
+#include "db_config.h"
+
+#ifndef lint
+static const char revid[] = "$Id: rep_method.c,v 1.78 2002/09/10 12:58:07 bostic Exp $";
+#endif /* not lint */
+
+#ifndef NO_SYSTEM_INCLUDES
+#include <sys/types.h>
+
+#ifdef HAVE_RPC
+#include <rpc/rpc.h>
+#endif
+
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#endif
+
+#include "db_int.h"
+#include "dbinc/db_page.h"
+#include "dbinc/db_am.h"
+#include "dbinc/log.h"
+#include "dbinc/rep.h"
+#include "dbinc/txn.h"
+
+#ifdef HAVE_RPC
+#include "dbinc_auto/db_server.h"
+#include "dbinc_auto/rpc_client_ext.h"
+#endif
+
+static int __rep_abort_prepared __P((DB_ENV *));
+static int __rep_bt_cmp __P((DB *, const DBT *, const DBT *));
+static int __rep_client_dbinit __P((DB_ENV *, int));
+static int __rep_elect __P((DB_ENV *, int, int, u_int32_t, int *));
+static int __rep_elect_init __P((DB_ENV *, DB_LSN *, int, int, int, int *));
+static int __rep_flush __P((DB_ENV *));
+static int __rep_restore_prepared __P((DB_ENV *));
+static int __rep_set_limit __P((DB_ENV *, u_int32_t, u_int32_t));
+static int __rep_set_request __P((DB_ENV *, u_int32_t, u_int32_t));
+static int __rep_set_rep_transport __P((DB_ENV *, int,
+ int (*)(DB_ENV *, const DBT *, const DBT *, int, u_int32_t)));
+static int __rep_start __P((DB_ENV *, DBT *, u_int32_t));
+static int __rep_stat __P((DB_ENV *, DB_REP_STAT **, u_int32_t));
+static int __rep_wait __P((DB_ENV *, u_int32_t, int *, u_int32_t));
+
+/*
+ * __rep_dbenv_create --
+ * Replication-specific initialization of the DB_ENV structure.
+ *
+ * PUBLIC: int __rep_dbenv_create __P((DB_ENV *));
+ */
+int
+__rep_dbenv_create(dbenv)
+ DB_ENV *dbenv;
+{
+ DB_REP *db_rep;
+ int ret;
+
+#ifdef HAVE_RPC
+ if (F_ISSET(dbenv, DB_ENV_RPCCLIENT)) {
+ COMPQUIET(db_rep, NULL);
+ COMPQUIET(ret, 0);
+ dbenv->rep_elect = __dbcl_rep_elect;
+ dbenv->rep_flush = __dbcl_rep_flush;
+ dbenv->rep_process_message = __dbcl_rep_process_message;
+ dbenv->rep_start = __dbcl_rep_start;
+ dbenv->rep_stat = __dbcl_rep_stat;
+ dbenv->set_rep_limit = __dbcl_rep_set_limit;
+ dbenv->set_rep_request = __dbcl_rep_set_request;
+ dbenv->set_rep_transport = __dbcl_rep_set_rep_transport;
+
+ } else
+#endif
+ {
+ dbenv->rep_elect = __rep_elect;
+ dbenv->rep_flush = __rep_flush;
+ dbenv->rep_process_message = __rep_process_message;
+ dbenv->rep_start = __rep_start;
+ dbenv->rep_stat = __rep_stat;
+ dbenv->set_rep_limit = __rep_set_limit;
+ dbenv->set_rep_request = __rep_set_request;
+ dbenv->set_rep_transport = __rep_set_rep_transport;
+ /*
+ * !!!
+ * Our caller has not yet had the opportunity to reset the panic
+ * state or turn off mutex locking, and so we can neither check
+ * the panic state or acquire a mutex in the DB_ENV create path.
+ */
+
+ if ((ret = __os_calloc(dbenv, 1, sizeof(DB_REP), &db_rep)) != 0)
+ return (ret);
+ dbenv->rep_handle = db_rep;
+
+ /* Initialize the per-process replication structure. */
+ db_rep->rep_send = NULL;
+ }
+
+ return (0);
+}
+
+/*
+ * __rep_start --
+ * Become a master or client, and start sending messages to participate
+ * in the replication environment. Must be called after the environment
+ * is open.
+ */
+static int
+__rep_start(dbenv, dbt, flags)
+ DB_ENV *dbenv;
+ DBT *dbt;
+ u_int32_t flags;
+{
+ DB_LOG *dblp;
+ DB_LSN lsn;
+ DB_REP *db_rep;
+ REP *rep;
+ int announce, init_db, redo_prepared, ret;
+
+ PANIC_CHECK(dbenv);
+ ENV_ILLEGAL_BEFORE_OPEN(dbenv, "rep_start");
+ ENV_REQUIRES_CONFIG(dbenv, dbenv->tx_handle, "rep_stat", DB_INIT_TXN);
+
+ db_rep = dbenv->rep_handle;
+ rep = db_rep->region;
+
+ if ((ret = __db_fchk(dbenv, "DB_ENV->rep_start", flags,
+ DB_REP_CLIENT | DB_REP_LOGSONLY | DB_REP_MASTER)) != 0)
+ return (ret);
+
+ /* Exactly one of CLIENT and MASTER must be specified. */
+ if ((ret = __db_fcchk(dbenv,
+ "DB_ENV->rep_start", flags, DB_REP_CLIENT, DB_REP_MASTER)) != 0)
+ return (ret);
+ if (!LF_ISSET(DB_REP_CLIENT | DB_REP_MASTER | DB_REP_LOGSONLY)) {
+ __db_err(dbenv,
+ "DB_ENV->rep_start: replication mode must be specified");
+ return (EINVAL);
+ }
+
+ /* Masters can't be logs-only. */
+ if ((ret = __db_fcchk(dbenv,
+ "DB_ENV->rep_start", flags, DB_REP_LOGSONLY, DB_REP_MASTER)) != 0)
+ return (ret);
+
+ /* We need a transport function. */
+ if (db_rep->rep_send == NULL) {
+ __db_err(dbenv,
+ "DB_ENV->set_rep_transport must be called before DB_ENV->rep_start");
+ return (EINVAL);
+ }
+
+ /* We'd better not have any logged files open if we are a client. */
+ if (LF_ISSET(DB_REP_CLIENT) && (ret = __dbreg_nofiles(dbenv)) != 0) {
+ __db_err(dbenv, "DB_ENV->rep_start called with open files");
+ return (ret);
+ }
+
+ MUTEX_LOCK(dbenv, db_rep->mutexp);
+ if (rep->eid == DB_EID_INVALID)
+ rep->eid = dbenv->rep_eid;
+
+ if (LF_ISSET(DB_REP_MASTER)) {
+ if (F_ISSET(dbenv, DB_ENV_REP_CLIENT)) {
+ /*
+ * If we're upgrading from having been a client,
+ * preclose, so that we close our temporary database.
+ *
+ * Do not close files that we may have opened while
+ * doing a rep_apply; they'll get closed when we
+ * finally close the environment, but for now, leave
+ * them open, as we don't want to recycle their
+ * fileids, and we may need the handles again if
+ * we become a client and the original master
+ * that opened them becomes a master again.
+ */
+ if ((ret = __rep_preclose(dbenv, 0)) != 0)
+ return (ret);
+
+ /*
+ * Now write a __txn_recycle record so that
+ * clients don't get confused with our txnids
+ * and txnids of previous masters.
+ */
+ F_CLR(dbenv, DB_ENV_REP_CLIENT);
+ if ((ret = __txn_reset(dbenv)) != 0)
+ return (ret);
+ }
+
+ redo_prepared = 0;
+ if (!F_ISSET(rep, REP_F_MASTER)) {
+ /* Master is not yet set. */
+ if (F_ISSET(rep, REP_ISCLIENT)) {
+ F_CLR(rep, REP_ISCLIENT);
+ rep->gen = ++rep->w_gen;
+ redo_prepared = 1;
+ } else if (rep->gen == 0)
+ rep->gen = 1;
+ }
+
+ F_SET(rep, REP_F_MASTER);
+ F_SET(dbenv, DB_ENV_REP_MASTER);
+ MUTEX_UNLOCK(dbenv, db_rep->mutexp);
+ dblp = (DB_LOG *)dbenv->lg_handle;
+ R_LOCK(dbenv, &dblp->reginfo);
+ lsn = ((LOG *)dblp->reginfo.primary)->lsn;
+ R_UNLOCK(dbenv, &dblp->reginfo);
+
+ /*
+ * Send the NEWMASTER message, then restore prepared txns
+ * if and only if we just upgraded from being a client.
+ */
+ if ((ret = __rep_send_message(dbenv,
+ DB_EID_BROADCAST, REP_NEWMASTER, &lsn, NULL, 0)) == 0 &&
+ redo_prepared)
+ ret = __rep_restore_prepared(dbenv);
+ } else {
+ F_CLR(dbenv, DB_ENV_REP_MASTER);
+ F_SET(dbenv, DB_ENV_REP_CLIENT);
+ if (LF_ISSET(DB_REP_LOGSONLY))
+ F_SET(dbenv, DB_ENV_REP_LOGSONLY);
+
+ announce = !F_ISSET(rep, REP_ISCLIENT) ||
+ rep->master_id == DB_EID_INVALID;
+ init_db = 0;
+ if (!F_ISSET(rep, REP_ISCLIENT)) {
+ F_CLR(rep, REP_F_MASTER);
+ if (LF_ISSET(DB_REP_LOGSONLY))
+ F_SET(rep, REP_F_LOGSONLY);
+ else
+ F_SET(rep, REP_F_UPGRADE);
+
+ /*
+ * We initialize the client's generation number to 0.
+ * Upon startup, it looks for a master and updates the
+ * generation number as necessary, exactly as it does
+ * during normal operation and a master failure.
+ */
+ rep->gen = 0;
+ rep->master_id = DB_EID_INVALID;
+ init_db = 1;
+ }
+ MUTEX_UNLOCK(dbenv, db_rep->mutexp);
+
+ /*
+ * Abort any prepared transactions that were restored
+ * by recovery. We won't be able to create any txns of
+ * our own until they're resolved, but we can't resolve
+ * them ourselves; the master has to. If any get
+ * resolved as commits, we'll redo them when commit
+ * records come in. Aborts will simply be ignored.
+ */
+ if ((ret = __rep_abort_prepared(dbenv)) != 0)
+ return (ret);
+
+ if ((ret = __rep_client_dbinit(dbenv, init_db)) != 0)
+ return (ret);
+
+ /*
+ * If this client created a newly replicated environment,
+ * then announce the existence of this client. The master
+ * should respond with a message that will tell this client
+ * the current generation number and the current LSN. This
+ * will allow the client to either perform recovery or
+ * simply join in.
+ */
+ if (announce)
+ ret = __rep_send_message(dbenv,
+ DB_EID_BROADCAST, REP_NEWCLIENT, NULL, dbt, 0);
+ }
+ return (ret);
+}
+
+/*
+ * __rep_client_dbinit --
+ *
+ * Initialize the LSN database on the client side. This is called from the
+ * client initialization code. The startup flag value indicates if
+ * this is the first thread/process starting up and therefore should create
+ * the LSN database. This routine must be called once by each process acting
+ * as a client.
+ */
+static int
+__rep_client_dbinit(dbenv, startup)
+ DB_ENV *dbenv;
+ int startup;
+{
+ DB_REP *db_rep;
+ DB *dbp;
+ int ret, t_ret;
+ u_int32_t flags;
+
+ PANIC_CHECK(dbenv);
+ db_rep = dbenv->rep_handle;
+ dbp = NULL;
+
+#define REPDBNAME "__db.rep.db"
+
+ /* Check if this has already been called on this environment. */
+ if (db_rep->rep_db != NULL)
+ return (0);
+
+ MUTEX_LOCK(dbenv, db_rep->db_mutexp);
+
+ if (startup) {
+ if ((ret = db_create(&dbp, dbenv, 0)) != 0)
+ goto err;
+ /*
+ * Ignore errors, because if the file doesn't exist, this
+ * is perfectly OK.
+ */
+ (void)dbp->remove(dbp, REPDBNAME, NULL, 0);
+ }
+
+ if ((ret = db_create(&dbp, dbenv, 0)) != 0)
+ goto err;
+ if ((ret = dbp->set_bt_compare(dbp, __rep_bt_cmp)) != 0)
+ goto err;
+
+ /* Allow writes to this database on a client. */
+ F_SET(dbp, DB_AM_CL_WRITER);
+
+ flags = (F_ISSET(dbenv, DB_ENV_THREAD) ? DB_THREAD : 0) |
+ (startup ? DB_CREATE : 0);
+ if ((ret = dbp->open(dbp, NULL,
+ "__db.rep.db", NULL, DB_BTREE, flags, 0)) != 0)
+ goto err;
+
+ db_rep->rep_db = dbp;
+
+ if (0) {
+err: if (dbp != NULL &&
+ (t_ret = dbp->close(dbp, DB_NOSYNC)) != 0 && ret == 0)
+ ret = t_ret;
+ db_rep->rep_db = NULL;
+ }
+
+ MUTEX_UNLOCK(dbenv, db_rep->db_mutexp);
+
+ return (ret);
+}
+
+/*
+ * __rep_bt_cmp --
+ *
+ * Comparison function for the LSN table. We use the entire control
+ * structure as a key (for simplicity, so we don't have to merge the
+ * other fields in the control with the data field), but really only
+ * care about the LSNs.
+ */
+static int
+__rep_bt_cmp(dbp, dbt1, dbt2)
+ DB *dbp;
+ const DBT *dbt1, *dbt2;
+{
+ DB_LSN lsn1, lsn2;
+ REP_CONTROL *rp1, *rp2;
+
+ COMPQUIET(dbp, NULL);
+
+ rp1 = dbt1->data;
+ rp2 = dbt2->data;
+
+ __ua_memcpy(&lsn1, &rp1->lsn, sizeof(DB_LSN));
+ __ua_memcpy(&lsn2, &rp2->lsn, sizeof(DB_LSN));
+
+ if (lsn1.file > lsn2.file)
+ return (1);
+
+ if (lsn1.file < lsn2.file)
+ return (-1);
+
+ if (lsn1.offset > lsn2.offset)
+ return (1);
+
+ if (lsn1.offset < lsn2.offset)
+ return (-1);
+
+ return (0);
+}
+
+/*
+ * __rep_abort_prepared --
+ * Abort any prepared transactions that recovery restored.
+ *
+ * This is used by clients that have just run recovery, since
+ * they cannot/should not call txn_recover and handle prepared transactions
+ * themselves.
+ */
+static int
+__rep_abort_prepared(dbenv)
+ DB_ENV *dbenv;
+{
+#define PREPLISTSIZE 50
+ DB_PREPLIST prep[PREPLISTSIZE], *p;
+ DB_TXNMGR *mgr;
+ DB_TXNREGION *region;
+ int do_aborts, ret;
+ long count, i;
+ u_int32_t op;
+
+ mgr = dbenv->tx_handle;
+ region = mgr->reginfo.primary;
+
+ do_aborts = 0;
+ R_LOCK(dbenv, &mgr->reginfo);
+ if (region->stat.st_nrestores != 0)
+ do_aborts = 1;
+ R_UNLOCK(dbenv, &mgr->reginfo);
+
+ if (do_aborts) {
+ op = DB_FIRST;
+ do {
+ if ((ret = dbenv->txn_recover(dbenv,
+ prep, PREPLISTSIZE, &count, op)) != 0)
+ return (ret);
+ for (i = 0; i < count; i++) {
+ p = &prep[i];
+ if ((ret = p->txn->abort(p->txn)) != 0)
+ return (ret);
+ }
+ op = DB_NEXT;
+ } while (count == PREPLISTSIZE);
+ }
+
+ return (0);
+}
+
+/*
+ * __rep_restore_prepared --
+ * Restore to a prepared state any prepared but not yet committed
+ * transactions.
+ *
+ * This performs, in effect, a "mini-recovery"; it is called from
+ * __rep_start by newly upgraded masters. There may be transactions that an
+ * old master prepared but did not resolve, which we need to restore to an
+ * active state.
+ */
+static int
+__rep_restore_prepared(dbenv)
+ DB_ENV *dbenv;
+{
+ DB_LOGC *logc;
+ DB_LSN ckp_lsn, lsn;
+ DBT rec;
+ __txn_ckp_args *ckp_args;
+ __txn_regop_args *regop_args;
+ __txn_xa_regop_args *prep_args;
+ int ret, t_ret;
+ u_int32_t hi_txn, low_txn, rectype;
+ void *txninfo;
+
+ txninfo = NULL;
+ ckp_args = NULL;
+ prep_args = NULL;
+ regop_args = NULL;
+ ZERO_LSN(ckp_lsn);
+ ZERO_LSN(lsn);
+
+ if ((ret = dbenv->log_cursor(dbenv, &logc, 0)) != 0)
+ return (ret);
+
+ /*
+ * We need to consider the set of records between the most recent
+ * checkpoint LSN and the end of the log; any txn in that
+ * range, and only txns in that range, could still have been
+ * active, and thus prepared but not yet committed (PBNYC),
+ * when the old master died.
+ *
+ * Find the most recent checkpoint LSN, and get the record there.
+ * If there is no checkpoint in the log, start off by getting
+ * the very first record in the log instead.
+ */
+ memset(&rec, 0, sizeof(DBT));
+ if ((ret = __txn_getckp(dbenv, &lsn)) == 0) {
+ if ((ret = logc->get(logc, &lsn, &rec, DB_SET)) != 0) {
+ __db_err(dbenv,
+ "Checkpoint record at LSN [%lu][%lu] not found",
+ (u_long)lsn.file, (u_long)lsn.offset);
+ goto err;
+ }
+
+ if ((ret = __txn_ckp_read(dbenv, rec.data, &ckp_args)) != 0) {
+ __db_err(dbenv,
+ "Invalid checkpoint record at [%lu][%lu]",
+ (u_long)lsn.file, (u_long)lsn.offset);
+ goto err;
+ }
+
+ ckp_lsn = ckp_args->ckp_lsn;
+ __os_free(dbenv, ckp_args);
+
+ if ((ret = logc->get(logc, &ckp_lsn, &rec, DB_SET)) != 0) {
+ __db_err(dbenv,
+ "Checkpoint LSN record [%lu][%lu] not found",
+ (u_long)ckp_lsn.file, (u_long)ckp_lsn.offset);
+ goto err;
+ }
+ } else if ((ret = logc->get(logc, &lsn, &rec, DB_FIRST)) != 0) {
+ if (ret == DB_NOTFOUND) {
+ /* An empty log means no PBNYC txns. */
+ ret = 0;
+ goto done;
+ }
+ __db_err(dbenv, "Attempt to get first log record failed");
+ goto err;
+ }
+
+ /*
+ * We use the same txnlist infrastructure that recovery does;
+ * it demands an estimate of the high and low txnids for
+ * initialization.
+ *
+ * First, the low txnid.
+ */
+ do {
+ /* txnid is after rectype, which is a u_int32. */
+ memcpy(&low_txn,
+ (u_int8_t *)rec.data + sizeof(u_int32_t), sizeof(low_txn));
+ if (low_txn != 0)
+ break;
+ } while ((ret = logc->get(logc, &lsn, &rec, DB_NEXT)) == 0);
+
+ /* If there are no txns, there are no PBNYC txns. */
+ if (ret == DB_NOTFOUND) {
+ ret = 0;
+ goto done;
+ } else if (ret != 0)
+ goto err;
+
+ /* Now, the high txnid. */
+ if ((ret = logc->get(logc, &lsn, &rec, DB_LAST)) != 0) {
+ /*
+ * Note that DB_NOTFOUND is unacceptable here because we
+ * had to have looked at some log record to get this far.
+ */
+ __db_err(dbenv, "Final log record not found");
+ goto err;
+ }
+ do {
+ /* txnid is after rectype, which is a u_int32. */
+ memcpy(&hi_txn,
+ (u_int8_t *)rec.data + sizeof(u_int32_t), sizeof(hi_txn));
+ if (hi_txn != 0)
+ break;
+ } while ((ret = logc->get(logc, &lsn, &rec, DB_PREV)) == 0);
+ if (ret == DB_NOTFOUND) {
+ ret = 0;
+ goto done;
+ } else if (ret != 0)
+ goto err;
+
+ /* We have a high and low txnid. Initialise the txn list. */
+ if ((ret =
+ __db_txnlist_init(dbenv, low_txn, hi_txn, NULL, &txninfo)) != 0)
+ goto err;
+
+ /*
+ * Now, walk backward from the end of the log to ckp_lsn. Any
+ * prepares that we hit without first hitting a commit or
+ * abort belong to PBNYC txns, and we need to apply them and
+ * restore them to a prepared state.
+ *
+ * Note that we wind up applying transactions out of order.
+ * Since all PBNYC txns still held locks on the old master and
+ * were isolated, this should be safe.
+ */
+ for (ret = logc->get(logc, &lsn, &rec, DB_LAST);
+ ret == 0 && log_compare(&lsn, &ckp_lsn) > 0;
+ ret = logc->get(logc, &lsn, &rec, DB_PREV)) {
+ memcpy(&rectype, rec.data, sizeof(rectype));
+ switch (rectype) {
+ case DB___txn_regop:
+ /*
+ * It's a commit or abort--but we don't care
+ * which! Just add it to the list of txns
+ * that are resolved.
+ */
+ if ((ret = __txn_regop_read(dbenv, rec.data,
+ &regop_args)) != 0)
+ goto err;
+
+ ret = __db_txnlist_find(dbenv,
+ txninfo, regop_args->txnid->txnid);
+ if (ret == DB_NOTFOUND)
+ ret = __db_txnlist_add(dbenv, txninfo,
+ regop_args->txnid->txnid,
+ regop_args->opcode, &lsn);
+ __os_free(dbenv, regop_args);
+ break;
+ case DB___txn_xa_regop:
+ /*
+ * It's a prepare. If we haven't put the
+ * txn on our list yet, it hasn't been
+ * resolved, so apply and restore it.
+ */
+ if ((ret = __txn_xa_regop_read(dbenv, rec.data,
+ &prep_args)) != 0)
+ goto err;
+ ret = __db_txnlist_find(dbenv, txninfo,
+ prep_args->txnid->txnid);
+ if (ret == DB_NOTFOUND)
+ if ((ret = __rep_process_txn(dbenv, &rec)) == 0)
+ ret = __txn_restore_txn(dbenv,
+ &lsn, prep_args);
+ __os_free(dbenv, prep_args);
+ break;
+ default:
+ continue;
+ }
+ }
+
+ /* It's not an error to have hit the beginning of the log. */
+ if (ret == DB_NOTFOUND)
+ ret = 0;
+
+done:
+err: t_ret = logc->close(logc, 0);
+
+ if (txninfo != NULL)
+ __db_txnlist_end(dbenv, txninfo);
+
+ return (ret == 0 ? t_ret : ret);
+}
+
+/*
+ * __rep_set_limit --
+ * Set a limit on the amount of data that will be sent during a single
+ * invocation of __rep_process_message.
+ */
+static int
+__rep_set_limit(dbenv, gbytes, bytes)
+ DB_ENV *dbenv;
+ u_int32_t gbytes;
+ u_int32_t bytes;
+{
+ DB_REP *db_rep;
+ REP *rep;
+
+ PANIC_CHECK(dbenv);
+
+ if ((db_rep = dbenv->rep_handle) == NULL) {
+ __db_err(dbenv,
+ "DB_ENV->set_rep_limit: database environment not properly initialized");
+ return (__db_panic(dbenv, EINVAL));
+ }
+ rep = db_rep->region;
+ MUTEX_LOCK(dbenv, db_rep->mutexp);
+ if (bytes > GIGABYTE) {
+ gbytes += bytes / GIGABYTE;
+ bytes = bytes % GIGABYTE;
+ }
+ rep->gbytes = gbytes;
+ rep->bytes = bytes;
+ MUTEX_UNLOCK(dbenv, db_rep->mutexp);
+
+ return (0);
+}
+
+/*
+ * __rep_set_request --
+ * Set the minimum and maximum number of log records that we wait
+ * before retransmitting.
+ * UNDOCUMENTED.
+ */
+static int
+__rep_set_request(dbenv, min, max)
+ DB_ENV *dbenv;
+ u_int32_t min;
+ u_int32_t max;
+{
+ LOG *lp;
+ DB_LOG *dblp;
+ DB_REP *db_rep;
+ REP *rep;
+
+ PANIC_CHECK(dbenv);
+
+ if ((db_rep = dbenv->rep_handle) == NULL) {
+ __db_err(dbenv,
+ "DB_ENV->set_rep_request: database environment not properly initialized");
+ return (__db_panic(dbenv, EINVAL));
+ }
+ rep = db_rep->region;
+ MUTEX_LOCK(dbenv, db_rep->mutexp);
+ rep->request_gap = min;
+ rep->max_gap = max;
+ MUTEX_UNLOCK(dbenv, db_rep->mutexp);
+ dblp = dbenv->lg_handle;
+ if (dblp != NULL && (lp = dblp->reginfo.primary) != NULL) {
+ R_LOCK(dbenv, &dblp->reginfo);
+ lp->wait_recs = 0;
+ lp->rcvd_recs = 0;
+ R_UNLOCK(dbenv, &dblp->reginfo);
+ }
+
+ return (0);
+}
+
+/*
+ * __rep_set_transport --
+ * Set the transport function for replication.
+ */
+static int
+__rep_set_rep_transport(dbenv, eid, f_send)
+ DB_ENV *dbenv;
+ int eid;
+ int (*f_send) __P((DB_ENV *, const DBT *, const DBT *, int, u_int32_t));
+{
+ DB_REP *db_rep;
+
+ PANIC_CHECK(dbenv);
+
+ if ((db_rep = dbenv->rep_handle) == NULL) {
+ __db_err(dbenv,
+ "DB_ENV->set_rep_transport: database environment not properly initialized");
+ return (__db_panic(dbenv, EINVAL));
+ }
+
+ if (f_send == NULL) {
+ __db_err(dbenv,
+ "DB_ENV->set_rep_transport: no send function specified");
+ return (EINVAL);
+ }
+
+ if (eid < 0) {
+ __db_err(dbenv,
+ "DB_ENV->set_rep_transport: eid must be greater than or equal to 0");
+ return (EINVAL);
+ }
+
+ db_rep->rep_send = f_send;
+
+ dbenv->rep_eid = eid;
+ return (0);
+}
+
+/*
+ * __rep_elect --
+ * Called after master failure to hold/participate in an election for
+ * a new master.
+ */
+static int
+__rep_elect(dbenv, nsites, priority, timeout, eidp)
+ DB_ENV *dbenv;
+ int nsites, priority;
+ u_int32_t timeout;
+ int *eidp;
+{
+ DB_LOG *dblp;
+ DB_LSN lsn;
+ DB_REP *db_rep;
+ REP *rep;
+ int in_progress, ret, send_vote, tiebreaker;
+ u_int32_t pid, sec, usec;
+
+ PANIC_CHECK(dbenv);
+ ENV_REQUIRES_CONFIG(dbenv, dbenv->tx_handle, "rep_elect", DB_INIT_TXN);
+
+ /* Error checking. */
+ if (nsites <= 0) {
+ __db_err(dbenv,
+ "DB_ENV->rep_elect: nsites must be greater than 0");
+ return (EINVAL);
+ }
+ if (priority < 0) {
+ __db_err(dbenv,
+ "DB_ENV->rep_elect: priority may not be negative");
+ return (EINVAL);
+ }
+
+ db_rep = dbenv->rep_handle;
+ rep = db_rep->region;
+ dblp = dbenv->lg_handle;
+
+ R_LOCK(dbenv, &dblp->reginfo);
+ lsn = ((LOG *)dblp->reginfo.primary)->lsn;
+ R_UNLOCK(dbenv, &dblp->reginfo);
+
+ /* Generate a randomized tiebreaker value. */
+ __os_id(&pid);
+ if ((ret = __os_clock(dbenv, &sec, &usec)) != 0)
+ return (ret);
+ tiebreaker = pid ^ sec ^ usec ^ (u_int)rand() ^ P_TO_UINT32(&pid);
+
+ if ((ret = __rep_elect_init(dbenv,
+ &lsn, nsites, priority, tiebreaker, &in_progress)) != 0) {
+ if (ret == DB_REP_NEWMASTER) {
+ ret = 0;
+ *eidp = dbenv->rep_eid;
+ }
+ return (ret);
+ }
+
+ if (!in_progress) {
+#ifdef DIAGNOSTIC
+ if (FLD_ISSET(dbenv->verbose, DB_VERB_REPLICATION))
+ __db_err(dbenv, "Beginning an election");
+#endif
+ if ((ret = __rep_send_message(dbenv,
+ DB_EID_BROADCAST, REP_ELECT, NULL, NULL, 0)) != 0)
+ goto err;
+ DB_ENV_TEST_RECOVERY(dbenv, DB_TEST_ELECTSEND, ret, NULL);
+ }
+
+ /* Now send vote */
+ if ((ret =
+ __rep_send_vote(dbenv, &lsn, nsites, priority, tiebreaker)) != 0)
+ goto err;
+ DB_ENV_TEST_RECOVERY(dbenv, DB_TEST_ELECTVOTE1, ret, NULL);
+
+ ret = __rep_wait(dbenv, timeout, eidp, REP_F_EPHASE1);
+ DB_ENV_TEST_RECOVERY(dbenv, DB_TEST_ELECTWAIT1, ret, NULL);
+ switch (ret) {
+ case 0:
+ /* Check if election complete or phase complete. */
+ if (*eidp != DB_EID_INVALID)
+ return (0);
+ goto phase2;
+ case DB_TIMEOUT:
+ break;
+ default:
+ goto err;
+ }
+ /*
+ * If we got here, we haven't heard from everyone, but we've
+ * run out of time, so it's time to decide if we have enough
+ * votes to pick a winner and if so, to send out a vote to
+ * the winner.
+ */
+ MUTEX_LOCK(dbenv, db_rep->mutexp);
+ send_vote = DB_EID_INVALID;
+ if (rep->sites > rep->nsites / 2) {
+ /* We think we've seen enough to cast a vote. */
+ send_vote = rep->winner;
+ if (rep->winner == rep->eid)
+ rep->votes++;
+ F_CLR(rep, REP_F_EPHASE1);
+ F_SET(rep, REP_F_EPHASE2);
+ }
+ MUTEX_UNLOCK(dbenv, db_rep->mutexp);
+ if (send_vote == DB_EID_INVALID) {
+ /* We do not have enough votes to elect. */
+#ifdef DIAGNOSTIC
+ if (FLD_ISSET(dbenv->verbose, DB_VERB_REPLICATION))
+ __db_err(dbenv,
+ "Not enough votes to elect: received %d of %d",
+ rep->sites, rep->nsites);
+#endif
+ ret = DB_REP_UNAVAIL;
+ goto err;
+
+ }
+#ifdef DIAGNOSTIC
+ if (FLD_ISSET(dbenv->verbose, DB_VERB_REPLICATION) &&
+ send_vote != rep->eid)
+ __db_err(dbenv, "Sending vote");
+#endif
+
+ if (send_vote != rep->eid && (ret = __rep_send_message(dbenv,
+ send_vote, REP_VOTE2, NULL, NULL, 0)) != 0)
+ goto err;
+ DB_ENV_TEST_RECOVERY(dbenv, DB_TEST_ELECTVOTE2, ret, NULL);
+
+phase2: ret = __rep_wait(dbenv, timeout, eidp, REP_F_EPHASE2);
+ DB_ENV_TEST_RECOVERY(dbenv, DB_TEST_ELECTWAIT2, ret, NULL);
+ switch (ret) {
+ case 0:
+ return (0);
+ case DB_TIMEOUT:
+ ret = DB_REP_UNAVAIL;
+ break;
+ default:
+ goto err;
+ }
+
+DB_TEST_RECOVERY_LABEL
+err: MUTEX_LOCK(dbenv, db_rep->mutexp);
+ ELECTION_DONE(rep);
+ MUTEX_UNLOCK(dbenv, db_rep->mutexp);
+
+#ifdef DIAGNOSTIC
+ if (FLD_ISSET(dbenv->verbose, DB_VERB_REPLICATION))
+ __db_err(dbenv, "Ended election with %d", ret);
+#endif
+ return (ret);
+}
+
+/*
+ * __rep_elect_init
+ * Initialize an election. Sets beginp non-zero if the election is
+ * already in progress; makes it 0 otherwise.
+ */
+static int
+__rep_elect_init(dbenv, lsnp, nsites, priority, tiebreaker, beginp)
+ DB_ENV *dbenv;
+ DB_LSN *lsnp;
+ int nsites, priority, tiebreaker, *beginp;
+{
+ DB_REP *db_rep;
+ REP *rep;
+ int ret, *tally;
+
+ db_rep = dbenv->rep_handle;
+ rep = db_rep->region;
+
+ ret = 0;
+
+ /* We may miscount, as we don't hold the replication mutex here. */
+ rep->stat.st_elections++;
+
+ /* If we are already a master; simply broadcast that fact and return. */
+ if (F_ISSET(dbenv, DB_ENV_REP_MASTER)) {
+ (void)__rep_send_message(dbenv,
+ DB_EID_BROADCAST, REP_NEWMASTER, lsnp, NULL, 0);
+ rep->stat.st_elections_won++;
+ return (DB_REP_NEWMASTER);
+ }
+
+ MUTEX_LOCK(dbenv, db_rep->mutexp);
+ *beginp = IN_ELECTION(rep);
+ if (!*beginp) {
+ /*
+ * Make sure that we always initialize all the election fields
+ * before putting ourselves in an election state. That means
+ * issuing calls that can fail (allocation) before setting all
+ * the variables.
+ */
+ if (nsites > rep->asites &&
+ (ret = __rep_grow_sites(dbenv, nsites)) != 0)
+ goto err;
+ DB_ENV_TEST_RECOVERY(dbenv, DB_TEST_ELECTINIT, ret, NULL);
+ rep->nsites = nsites;
+ rep->priority = priority;
+ rep->votes = 0;
+ rep->master_id = DB_EID_INVALID;
+ F_SET(rep, REP_F_EPHASE1);
+
+ /* We have always heard from ourselves. */
+ rep->sites = 1;
+ tally = R_ADDR((REGINFO *)dbenv->reginfo, rep->tally_off);
+ tally[0] = rep->eid;
+
+ if (priority != 0) {
+ /* Make ourselves the winner to start. */
+ rep->winner = rep->eid;
+ rep->w_priority = priority;
+ rep->w_gen = rep->gen;
+ rep->w_lsn = *lsnp;
+ rep->w_tiebreaker = tiebreaker;
+ } else {
+ rep->winner = DB_EID_INVALID;
+ rep->w_priority = 0;
+ rep->w_gen = 0;
+ ZERO_LSN(rep->w_lsn);
+ rep->w_tiebreaker = 0;
+ }
+ }
+DB_TEST_RECOVERY_LABEL
+err: MUTEX_UNLOCK(dbenv, db_rep->mutexp);
+ return (ret);
+}
+
+static int
+__rep_wait(dbenv, timeout, eidp, flags)
+ DB_ENV *dbenv;
+ u_int32_t timeout;
+ int *eidp;
+ u_int32_t flags;
+{
+ DB_REP *db_rep;
+ REP *rep;
+ int done, ret;
+ u_int32_t sleeptime;
+
+ done = 0;
+ db_rep = dbenv->rep_handle;
+ rep = db_rep->region;
+
+ /*
+ * The user specifies an overall timeout function, but checking
+ * is cheap and the timeout may be a generous upper bound.
+ * Sleep repeatedly for the smaller of .5s and timeout/10.
+ */
+ sleeptime = (timeout > 5000000) ? 500000 : timeout / 10;
+ if (sleeptime == 0)
+ sleeptime++;
+ while (timeout > 0) {
+ if ((ret = __os_sleep(dbenv, 0, sleeptime)) != 0)
+ return (ret);
+ MUTEX_LOCK(dbenv, db_rep->mutexp);
+ done = !F_ISSET(rep, flags) && rep->master_id != DB_EID_INVALID;
+
+ *eidp = rep->master_id;
+ MUTEX_UNLOCK(dbenv, db_rep->mutexp);
+
+ if (done)
+ return (0);
+
+ if (timeout > sleeptime)
+ timeout -= sleeptime;
+ else
+ timeout = 0;
+ }
+ return (DB_TIMEOUT);
+}
+
+/*
+ * __rep_flush --
+ * Re-push the last log record to all clients, in case they've lost
+ * messages and don't know it.
+ */
+static int
+__rep_flush(dbenv)
+ DB_ENV *dbenv;
+{
+ DBT rec;
+ DB_LOGC *logc;
+ DB_LSN lsn;
+ int ret, t_ret;
+
+ PANIC_CHECK(dbenv);
+ ENV_REQUIRES_CONFIG(dbenv, dbenv->tx_handle, "rep_stat", DB_INIT_TXN);
+
+ if ((ret = dbenv->log_cursor(dbenv, &logc, 0)) != 0)
+ return (ret);
+
+ memset(&rec, 0, sizeof(rec));
+ memset(&lsn, 0, sizeof(lsn));
+
+ if ((ret = logc->get(logc, &lsn, &rec, DB_LAST)) != 0)
+ goto err;
+
+ ret = __rep_send_message(dbenv,
+ DB_EID_BROADCAST, REP_LOG, &lsn, &rec, 0);
+
+err: if ((t_ret = logc->close(logc, 0)) != 0 && ret == 0)
+ ret = t_ret;
+ return (ret);
+}
+
+/*
+ * __rep_stat --
+ * Fetch replication statistics.
+ */
+static int
+__rep_stat(dbenv, statp, flags)
+ DB_ENV *dbenv;
+ DB_REP_STAT **statp;
+ u_int32_t flags;
+{
+ DB_LOG *dblp;
+ DB_REP *db_rep;
+ DB_REP_STAT *stats;
+ LOG *lp;
+ REP *rep;
+ u_int32_t queued;
+ int ret;
+
+ PANIC_CHECK(dbenv);
+ ENV_REQUIRES_CONFIG(dbenv, dbenv->tx_handle, "rep_stat", DB_INIT_TXN);
+
+ db_rep = dbenv->rep_handle;
+ rep = db_rep->region;
+ dblp = dbenv->lg_handle;
+ lp = dblp->reginfo.primary;
+
+ *statp = NULL;
+ if ((ret = __db_fchk(dbenv,
+ "DB_ENV->rep_stat", flags, DB_STAT_CLEAR)) != 0)
+ return (ret);
+
+ /* Allocate a stat struct to return to the user. */
+ if ((ret = __os_umalloc(dbenv, sizeof(DB_REP_STAT), &stats)) != 0)
+ return (ret);
+
+ MUTEX_LOCK(dbenv, db_rep->mutexp);
+ memcpy(stats, &rep->stat, sizeof(*stats));
+
+ /* Copy out election stats. */
+ if (IN_ELECTION(rep)) {
+ if (F_ISSET(rep, REP_F_EPHASE1))
+ stats->st_election_status = 1;
+ else if (F_ISSET(rep, REP_F_EPHASE2))
+ stats->st_election_status = 2;
+
+ stats->st_election_nsites = rep->sites;
+ stats->st_election_cur_winner = rep->winner;
+ stats->st_election_priority = rep->w_priority;
+ stats->st_election_gen = rep->w_gen;
+ stats->st_election_lsn = rep->w_lsn;
+ stats->st_election_votes = rep->votes;
+ stats->st_election_tiebreaker = rep->w_tiebreaker;
+ }
+
+ /* Copy out other info that's protected by the rep mutex. */
+ stats->st_env_id = rep->eid;
+ stats->st_env_priority = rep->priority;
+ stats->st_nsites = rep->nsites;
+ stats->st_master = rep->master_id;
+ stats->st_gen = rep->gen;
+
+ if (F_ISSET(rep, REP_F_MASTER))
+ stats->st_status = DB_REP_MASTER;
+ else if (F_ISSET(rep, REP_F_LOGSONLY))
+ stats->st_status = DB_REP_LOGSONLY;
+ else if (F_ISSET(rep, REP_F_UPGRADE))
+ stats->st_status = DB_REP_CLIENT;
+ else
+ stats->st_status = 0;
+
+ if (LF_ISSET(DB_STAT_CLEAR)) {
+ queued = rep->stat.st_log_queued;
+ memset(&rep->stat, 0, sizeof(rep->stat));
+ rep->stat.st_log_queued = rep->stat.st_log_queued_total =
+ rep->stat.st_log_queued_max = queued;
+ }
+ MUTEX_UNLOCK(dbenv, db_rep->mutexp);
+
+ /*
+ * Log-related replication info is stored in the log system and
+ * protected by the log region lock.
+ */
+ R_LOCK(dbenv, &dblp->reginfo);
+ if (F_ISSET(rep, REP_ISCLIENT)) {
+ stats->st_next_lsn = lp->ready_lsn;
+ stats->st_waiting_lsn = lp->waiting_lsn;
+ } else {
+ if (F_ISSET(rep, REP_F_MASTER))
+ stats->st_next_lsn = lp->lsn;
+ else
+ ZERO_LSN(stats->st_next_lsn);
+ ZERO_LSN(stats->st_waiting_lsn);
+ }
+ R_UNLOCK(dbenv, &dblp->reginfo);
+
+ *statp = stats;
+ return (0);
+}