mariadb/storage/innobase/buf/buf0flu.c
unknown 76de7d788c Apply the following InnoDB snapshots:
innodb-5.1-ss1318
innodb-5.1-ss1330
innodb-5.1-ss1332
innodb-5.1-ss1340

Fixes:
- Bug #21409: Incorrect result returned when in READ-COMMITTED with query_cache ON
  At low transaction isolation levels we let each consistent read set
  its own snapshot.

- Bug #23666: strange Innodb_row_lock_time_% values in show status; also millisecs wrong
  On Windows ut_usectime returns secs and usecs relative to the UNIX
  epoch (which is Jan, 1 1970).

- Bug #25494: LATEST DEADLOCK INFORMATION is not always cleared
  lock_deadlock_recursive(): When the search depth or length is exceeded,
  rewind lock_latest_err_file and display the two transactions at the
  point of aborting the search.

- Bug #25927: Foreign key with ON DELETE SET NULL on NOT NULL can crash server
  Prevent ALTER TABLE ... MODIFY ... NOT NULL on columns for which
  there is a foreign key constraint ON ... SET NULL.

- Bug #26835: Repeatable corruption of utf8-enabled tables inside InnoDB
  The bug could be reproduced as follows:

  Define a table so that the first column of the clustered index is
  a VARCHAR or a UTF-8 CHAR in a collation where sequences of bytes
  of differing length are considered equivalent.

  Insert and delete a record.  Before the delete-marked record is
  purged, insert another record whose first column is of different
  length but equivalent to the first record.  Under certain conditions,
  the insertion can be incorrectly performed as update-in-place.

  Likewise, an operation that could be done as update-in-place can
  unnecessarily be performed as delete and insert, but that would not
  cause corruption but merely degraded performance.


mysql-test/r/innodb.result:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1284:
  Merge changes from MySQL AB:
  
  ChangeSet
    2007/01/24 14:49:36+04:00 holyfoot@mysql.com 
    bug 22682 Test fails --without-geometry
    geometry dependent parts moved to proper .test files
  
  mysql-test/r/innodb.result
    2007/01/24 14:49:34+04:00 holyfoot@mysql.com +0 -2
    result fixed
  
  mysql-test/r/innodb_gis.result
    2007/01/24 14:49:34+04:00 holyfoot@mysql.com +2 -0
    result fixed
  
  mysql-test/t/innodb.test
    2007/01/24 14:49:34+04:00 holyfoot@mysql.com +0 -6
    HAVE_GEOMETRY dependent part moved to innodb_gis.test
  
  mysql-test/t/innodb_gis.test
    2007/01/24 14:49:35+04:00 holyfoot@mysql.com +6 -0
    HAVE_GEOMETRY dependent part moved here from innodb.test
  
  
  Revision r1186:
  dict_load_foreign(): Use a local variable instead of the 10-bit field
  foreign->n_fields in order to preserve ON UPDATE CASCADE and
  ON DELETE CASCADE flags.  For some reason, gcc does not warn about
  shifting a 10-bit field to right by 24 bits.  (Bug 24741)
  
  This bug was introduced while reducing the memory footprint of the
  InnoDB data dictionary (Bug 20877).
  
  innodb.test, innodb.result: Add a test case.
  
  
  Revision r1318:
  Add a test case for r1316 (Bug #25927).
  
  
  Revision r1340:
  innodb.test, innodb.result: Add test case for Bug #26835.
  The bug could be reproduced as follows:
  
  Define a table so that the first column of the clustered index is
  a VARCHAR or a UTF-8 CHAR in a collation where sequences of bytes
  of differing length are considered equivalent.
  
  Insert and delete a record.  Before the delete-marked record is
  purged, insert another record whose first column is of different
  length but equivalent to the first record.  Under certain conditions,
  the insertion can be incorrectly performed as update-in-place.
  
  Likewise, an operation that could be done as update-in-place can
  unnecessarily be performed as delete and insert, but that would not
  cause corruption but merely degraded performance.
mysql-test/t/innodb.test:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1284:
  Merge changes from MySQL AB:
  
  ChangeSet
    2007/01/24 14:49:36+04:00 holyfoot@mysql.com 
    bug 22682 Test fails --without-geometry
    geometry dependent parts moved to proper .test files
  
  mysql-test/r/innodb.result
    2007/01/24 14:49:34+04:00 holyfoot@mysql.com +0 -2
    result fixed
  
  mysql-test/r/innodb_gis.result
    2007/01/24 14:49:34+04:00 holyfoot@mysql.com +2 -0
    result fixed
  
  mysql-test/t/innodb.test
    2007/01/24 14:49:34+04:00 holyfoot@mysql.com +0 -6
    HAVE_GEOMETRY dependent part moved to innodb_gis.test
  
  mysql-test/t/innodb_gis.test
    2007/01/24 14:49:35+04:00 holyfoot@mysql.com +6 -0
    HAVE_GEOMETRY dependent part moved here from innodb.test
  
  
  Revision r1283:
  Merge changes from MySQL AB:
  
  ChangeSet
    2007/01/22 18:42:52+02:00 monty@mysql.com 
    Give warnings for unused objects
    Changed error message to be compatible with old error file
    Added new error message for new DUP_ENTRY syntax
  
  mysql-test/t/innodb.test
    2007/01/22 18:42:49+02:00 monty@mysql.com +14 -14
    Changed to use new error message
  
  
  Revision r1186:
  dict_load_foreign(): Use a local variable instead of the 10-bit field
  foreign->n_fields in order to preserve ON UPDATE CASCADE and
  ON DELETE CASCADE flags.  For some reason, gcc does not warn about
  shifting a 10-bit field to right by 24 bits.  (Bug 24741)
  
  This bug was introduced while reducing the memory footprint of the
  InnoDB data dictionary (Bug 20877).
  
  innodb.test, innodb.result: Add a test case.
  
  
  Revision r1318:
  Add a test case for r1316 (Bug #25927).
  
  
  Revision r1329:
  Merge changes from MySQL AB to mysql-test directives.
  The results are not affected.
  
  
  Revision r1340:
  innodb.test, innodb.result: Add test case for Bug #26835.
  The bug could be reproduced as follows:
  
  Define a table so that the first column of the clustered index is
  a VARCHAR or a UTF-8 CHAR in a collation where sequences of bytes
  of differing length are considered equivalent.
  
  Insert and delete a record.  Before the delete-marked record is
  purged, insert another record whose first column is of different
  length but equivalent to the first record.  Under certain conditions,
  the insertion can be incorrectly performed as update-in-place.
  
  Likewise, an operation that could be done as update-in-place can
  unnecessarily be performed as delete and insert, but that would not
  cause corruption but merely degraded performance.
storage/innobase/buf/buf0buf.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/buf/buf0flu.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/buf/buf0lru.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/dict/dict0boot.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/dict/dict0crea.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1324:
  Merge changes from MySQL AB:
  
  ChangeSet@1.2452, 2007-02-23 13:13:55+02:00, monty@mysql.com +177 -0
    Fixed compiler warnings
    ...
    Fixed compiler warnings detected on windows64
storage/innobase/dict/dict0dict.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1316:
  Prevent ALTER TABLE ... MODIFY ... NOT NULL on columns for which
  there is a foreign key constraint ON ... SET NULL.  (Bug #25927)
  
  dict_foreign_find_index(): Add paramettter check_null.
  
  dict_foreign_add_to_cache(): Do not allow ON DELETE SET NULL
  or ON UPDATE SET NULL if any of the referencing columns are declared NOT NULL.
  
  
  Revision r1324:
  Merge changes from MySQL AB:
  
  ChangeSet@1.2452, 2007-02-23 13:13:55+02:00, monty@mysql.com +177 -0
    Fixed compiler warnings
    ...
    Fixed compiler warnings detected on windows64
storage/innobase/dict/dict0load.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1186:
  dict_load_foreign(): Use a local variable instead of the 10-bit field
  foreign->n_fields in order to preserve ON UPDATE CASCADE and
  ON DELETE CASCADE flags.  For some reason, gcc does not warn about
  shifting a 10-bit field to right by 24 bits.  (Bug 24741)
  
  This bug was introduced while reducing the memory footprint of the
  InnoDB data dictionary (Bug 20877).
  
  innodb.test, innodb.result: Add a test case.
  
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1324:
  Merge changes from MySQL AB:
  
  ChangeSet@1.2452, 2007-02-23 13:13:55+02:00, monty@mysql.com +177 -0
    Fixed compiler warnings
    ...
    Fixed compiler warnings detected on windows64
storage/innobase/fil/fil0fil.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/fsp/fsp0fsp.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/ha/ha0ha.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/handler/ha_innodb.cc:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1204:
  Change this in ha_innobase:
  
   void*           innobase_prebuilt;
  
  to this:
  
   row_prebuilt_t* prebuilt;
  
  by introducing the typedef in ha_innodb.h, and remove all the now needless
  local variables and casts in ha_innodb.cc.
  
  Revision r1298:
  ha_innodb.cc: Remove all references to thd->ha_data[hton->slot].
  
  thd_to_trx(thd, hton): Accessor for getting the InnoDB trx object
  of a MySQL thread object and an InnoDB handlerton.
  
  
  Revision r1292:
  Remove the declarations of some global functions in ha_innodb.h and declare
  them static in ha_innodb.cc.  These functions are invoked via function
  pointers in handlerton.
  
  
  Revision r1300:
  ha_innodb.cc: Replace thd->tablespace_op with thd_tablespace_op(thd).
  Plugins must treat class THD as an opaque type.
  
  
  Revision r1198:
  Merge a change from MySQL AB:
  
  ChangeSet@1.2372, 2006-12-31 02:29:11+01:00, kent@mysql.com +79 -0
    Many files:
      Removed "MySQL Finland AB & TCX DataKonsult AB" from copyright header
      Adjusted year(s) in copyright header 
      Added GPL copyright text
  
  
  Revision r1271:
  Merge changes from MySQL AB:
  
  Rename some FIELD_TYPE_ constants to MYSQL_TYPE_.
  
  Change the scope of a type cast of two dividends.
  
  
  Revision r1299:
  ha_innodb.cc: Replace thd->in_lock_tables with thd_in_lock_tables(thd).
  Plugins must treat class THD as an opaque type.
  
  
  Revision r1201:
  Apply patch from MySQL:
  
   ChangeSet@1.2353, 2006-12-19 16:57:51-07:00, tsmith@siva.hindu.god +13 -0
     Added innodb_rollback_on_timeout option to restore the 4.1 
     InnoDB timeout behavior (Bug 24200)
  
  Revision r1322:
  ha_innodb.cc: Remove the unused innobase_repl_ variables.
  
  
  Revision r1324:
  Merge changes from MySQL AB:
  
  ChangeSet@1.2452, 2007-02-23 13:13:55+02:00, monty@mysql.com +177 -0
    Fixed compiler warnings
    ...
    Fixed compiler warnings detected on windows64
  
  
  Revision r1334:
  Fix for Bug# 21409. At low transaction isolation levels we let each 
  consistent read set its own snapshot
storage/innobase/handler/ha_innodb.h:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1204:
  Change this in ha_innobase:
  
   void*           innobase_prebuilt;
  
  to this:
  
   row_prebuilt_t* prebuilt;
  
  by introducing the typedef in ha_innodb.h, and remove all the now needless
  local variables and casts in ha_innodb.cc.
  
  Revision r1292:
  Remove the declarations of some global functions in ha_innodb.h and declare
  them static in ha_innodb.cc.  These functions are invoked via function
  pointers in handlerton.
  
  
  Revision r1198:
  Merge a change from MySQL AB:
  
  ChangeSet@1.2372, 2006-12-31 02:29:11+01:00, kent@mysql.com +79 -0
    Many files:
      Removed "MySQL Finland AB & TCX DataKonsult AB" from copyright header
      Adjusted year(s) in copyright header 
      Added GPL copyright text
  
  
  Revision r1201:
  Apply patch from MySQL:
  
   ChangeSet@1.2353, 2006-12-19 16:57:51-07:00, tsmith@siva.hindu.god +13 -0
     Added innodb_rollback_on_timeout option to restore the 4.1 
     InnoDB timeout behavior (Bug 24200)
storage/innobase/ibuf/ibuf0ibuf.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/buf0buf.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/buf0flu.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/dict0dict.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/ha0ha.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/lock0lock.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/log0log.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/mem0mem.h:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1241:
  Remove the unused function mem_strdupq().
storage/innobase/include/mem0mem.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1241:
  Remove the unused function mem_strdupq().
storage/innobase/include/rem0rec.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1338:
  rec_offs_nth_size(): Treat n==0 as a special case.  (Bug #26835)
storage/innobase/include/sync0rw.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/sync0sync.h:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1247:
  Rename mutex_enter_nowait to mutex_enter_nowait_func and add macro
  mutex_enter_nowait that supplies the default __FILE__ and __LINE__
  arguments. Adjust callers.
storage/innobase/include/sync0sync.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1294:
  Fixed inline asm code, it didn't work with GCC > ver 3.x.
  
  
  Revision r1244:
  Add ut_ad() debug assertions.
  
  UT_LIST_ADD_FIRST(), UT_LIST_ADD_LAST(), UT_LIST_INSERT_AFTER():
  Assert against some trivial cases of cyclic lists.
  
  mutex_enter_func(): Assert that the current thread is not holding the mutex.
  
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/trx0sys.ic:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/include/univ.i:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1285:
  Merge a change from MySQL AB:
  
  ChangeSet
    2006/10/26 15:41:47-04:00 iggy@amd64. 
    Post Merge Cleanup
  
  storage/innobase/include/univ.i
    2006/10/26 15:38:50-04:00 iggy@amd64. +9 -0
    Post Merge Cleanup
storage/innobase/include/ut0lst.h:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1244:
  Add ut_ad() debug assertions.
  
  UT_LIST_ADD_FIRST(), UT_LIST_ADD_LAST(), UT_LIST_INSERT_AFTER():
  Assert against some trivial cases of cyclic lists.
  
  mutex_enter_func(): Assert that the current thread is not holding the mutex.
storage/innobase/lock/lock0lock.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1330:
  lock_deadlock_recursive(): When the search depth or length is exceeded,
  rewind lock_latest_err_file and display the two transactions at the
  point of aborting the search.  (Bug #25494)
  
  
  Revision r1332:
  lock_deadlock_recursive(): When aborting the search, display a note
  regardless of start->undo_no.  Otherwise, aborted searches may show
  up as genuine deadlocks.  This mistake was made in r1330.
storage/innobase/log/log0log.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1247:
  Rename mutex_enter_nowait to mutex_enter_nowait_func and add macro
  mutex_enter_nowait that supplies the default __FILE__ and __LINE__
  arguments. Adjust callers.
storage/innobase/log/log0recv.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/mem/mem0pool.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/pars/pars0pars.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/que/que0que.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/read/read0read.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/row/row0mysql.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1201:
  Apply patch from MySQL:
  
   ChangeSet@1.2353, 2006-12-19 16:57:51-07:00, tsmith@siva.hindu.god +13 -0
     Added innodb_rollback_on_timeout option to restore the 4.1 
     InnoDB timeout behavior (Bug 24200)
  
  Revision r1324:
  Merge changes from MySQL AB:
  
  ChangeSet@1.2452, 2007-02-23 13:13:55+02:00, monty@mysql.com +177 -0
    Fixed compiler warnings
    ...
    Fixed compiler warnings detected on windows64
storage/innobase/row/row0vers.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/srv/srv0que.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/srv/srv0srv.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1262:
  Fix for Bug# 23666. On Windows ut_usectime returns secs 
  and usecs relative to the UNIX epoch (which is Jan, 1 1970).
  
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/sync/sync0rw.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1247:
  Rename mutex_enter_nowait to mutex_enter_nowait_func and add macro
  mutex_enter_nowait that supplies the default __FILE__ and __LINE__
  arguments. Adjust callers.
  
  Revision r1324:
  Merge changes from MySQL AB:
  
  ChangeSet@1.2452, 2007-02-23 13:13:55+02:00, monty@mysql.com +177 -0
    Fixed compiler warnings
    ...
    Fixed compiler warnings detected on windows64
storage/innobase/sync/sync0sync.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1247:
  Rename mutex_enter_nowait to mutex_enter_nowait_func and add macro
  mutex_enter_nowait that supplies the default __FILE__ and __LINE__
  arguments. Adjust callers.
storage/innobase/thr/thr0loc.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/trx/trx0purge.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/trx/trx0roll.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/trx/trx0rseg.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/trx/trx0sys.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/trx/trx0trx.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
  
  
  Revision r1324:
  Merge changes from MySQL AB:
  
  ChangeSet@1.2452, 2007-02-23 13:13:55+02:00, monty@mysql.com +177 -0
    Fixed compiler warnings
    ...
    Fixed compiler warnings detected on windows64
storage/innobase/trx/trx0undo.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/usr/usr0sess.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1242:
  Merge r1239 from
  branches/zip: Make mutex_own() work with UNIV_DEBUG, without UNIV_SYNC_DEBUG.
storage/innobase/ut/ut0ut.c:
  Apply the following InnoDB snapshots:
  innodb-5.1-ss1318
  innodb-5.1-ss1330
  innodb-5.1-ss1332
  innodb-5.1-ss1340
  
  Revision r1262:
  Fix for Bug# 23666. On Windows ut_usectime returns secs 
  and usecs relative to the UNIX epoch (which is Jan, 1 1970).
2007-03-22 15:59:35 -06:00

1115 lines
29 KiB
C

/******************************************************
The database buffer buf_pool flush algorithm
(c) 1995-2001 Innobase Oy
Created 11/11/1995 Heikki Tuuri
*******************************************************/
#include "buf0flu.h"
#ifdef UNIV_NONINL
#include "buf0flu.ic"
#include "trx0sys.h"
#endif
#include "ut0byte.h"
#include "ut0lst.h"
#include "page0page.h"
#include "fil0fil.h"
#include "buf0buf.h"
#include "buf0lru.h"
#include "buf0rea.h"
#include "ibuf0ibuf.h"
#include "log0log.h"
#include "os0file.h"
#include "trx0sys.h"
#include "srv0srv.h"
/* When flushed, dirty blocks are searched in neighborhoods of this size, and
flushed along with the original page. */
#define BUF_FLUSH_AREA ut_min(BUF_READ_AHEAD_AREA,\
buf_pool->curr_size / 16)
/**********************************************************************
Validates the flush list. */
static
ibool
buf_flush_validate_low(void);
/*========================*/
/* out: TRUE if ok */
/************************************************************************
Inserts a modified block into the flush list. */
void
buf_flush_insert_into_flush_list(
/*=============================*/
buf_block_t* block) /* in: block which is modified */
{
ut_ad(mutex_own(&(buf_pool->mutex)));
ut_a(block->state == BUF_BLOCK_FILE_PAGE);
ut_ad((UT_LIST_GET_FIRST(buf_pool->flush_list) == NULL)
|| (ut_dulint_cmp((UT_LIST_GET_FIRST(buf_pool->flush_list))
->oldest_modification,
block->oldest_modification) <= 0));
UT_LIST_ADD_FIRST(flush_list, buf_pool->flush_list, block);
ut_ad(buf_flush_validate_low());
}
/************************************************************************
Inserts a modified block into the flush list in the right sorted position.
This function is used by recovery, because there the modifications do not
necessarily come in the order of lsn's. */
void
buf_flush_insert_sorted_into_flush_list(
/*====================================*/
buf_block_t* block) /* in: block which is modified */
{
buf_block_t* prev_b;
buf_block_t* b;
ut_ad(mutex_own(&(buf_pool->mutex)));
prev_b = NULL;
b = UT_LIST_GET_FIRST(buf_pool->flush_list);
while (b && (ut_dulint_cmp(b->oldest_modification,
block->oldest_modification) > 0)) {
prev_b = b;
b = UT_LIST_GET_NEXT(flush_list, b);
}
if (prev_b == NULL) {
UT_LIST_ADD_FIRST(flush_list, buf_pool->flush_list, block);
} else {
UT_LIST_INSERT_AFTER(flush_list, buf_pool->flush_list, prev_b,
block);
}
ut_ad(buf_flush_validate_low());
}
/************************************************************************
Returns TRUE if the file page block is immediately suitable for replacement,
i.e., the transition FILE_PAGE => NOT_USED allowed. */
ibool
buf_flush_ready_for_replace(
/*========================*/
/* out: TRUE if can replace immediately */
buf_block_t* block) /* in: buffer control block, must be in state
BUF_BLOCK_FILE_PAGE and in the LRU list */
{
ut_ad(mutex_own(&(buf_pool->mutex)));
ut_ad(mutex_own(&block->mutex));
if (block->state != BUF_BLOCK_FILE_PAGE) {
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: Error: buffer block state %lu"
" in the LRU list!\n",
(ulong)block->state);
ut_print_buf(stderr, block, sizeof(buf_block_t));
return(FALSE);
}
if ((ut_dulint_cmp(block->oldest_modification, ut_dulint_zero) > 0)
|| (block->buf_fix_count != 0)
|| (block->io_fix != 0)) {
return(FALSE);
}
return(TRUE);
}
/************************************************************************
Returns TRUE if the block is modified and ready for flushing. */
UNIV_INLINE
ibool
buf_flush_ready_for_flush(
/*======================*/
/* out: TRUE if can flush immediately */
buf_block_t* block, /* in: buffer control block, must be in state
BUF_BLOCK_FILE_PAGE */
ulint flush_type)/* in: BUF_FLUSH_LRU or BUF_FLUSH_LIST */
{
ut_ad(mutex_own(&(buf_pool->mutex)));
ut_ad(mutex_own(&(block->mutex)));
ut_a(block->state == BUF_BLOCK_FILE_PAGE);
if ((ut_dulint_cmp(block->oldest_modification, ut_dulint_zero) > 0)
&& (block->io_fix == 0)) {
if (flush_type != BUF_FLUSH_LRU) {
return(TRUE);
} else if (block->buf_fix_count == 0) {
/* If we are flushing the LRU list, to avoid deadlocks
we require the block not to be bufferfixed, and hence
not latched. */
return(TRUE);
}
}
return(FALSE);
}
/************************************************************************
Updates the flush system data structures when a write is completed. */
void
buf_flush_write_complete(
/*=====================*/
buf_block_t* block) /* in: pointer to the block in question */
{
ut_ad(block);
#ifdef UNIV_SYNC_DEBUG
ut_ad(mutex_own(&(buf_pool->mutex)));
#endif /* UNIV_SYNC_DEBUG */
ut_a(block->state == BUF_BLOCK_FILE_PAGE);
block->oldest_modification = ut_dulint_zero;
UT_LIST_REMOVE(flush_list, buf_pool->flush_list, block);
ut_d(UT_LIST_VALIDATE(flush_list, buf_block_t, buf_pool->flush_list));
(buf_pool->n_flush[block->flush_type])--;
if (block->flush_type == BUF_FLUSH_LRU) {
/* Put the block to the end of the LRU list to wait to be
moved to the free list */
buf_LRU_make_block_old(block);
buf_pool->LRU_flush_ended++;
}
/* fprintf(stderr, "n pending flush %lu\n",
buf_pool->n_flush[block->flush_type]); */
if ((buf_pool->n_flush[block->flush_type] == 0)
&& (buf_pool->init_flush[block->flush_type] == FALSE)) {
/* The running flush batch has ended */
os_event_set(buf_pool->no_flush[block->flush_type]);
}
}
/************************************************************************
Flushes possible buffered writes from the doublewrite memory buffer to disk,
and also wakes up the aio thread if simulated aio is used. It is very
important to call this function after a batch of writes has been posted,
and also when we may have to wait for a page latch! Otherwise a deadlock
of threads can occur. */
static
void
buf_flush_buffered_writes(void)
/*===========================*/
{
buf_block_t* block;
byte* write_buf;
ulint len;
ulint len2;
ulint i;
if (!srv_use_doublewrite_buf || trx_doublewrite == NULL) {
os_aio_simulated_wake_handler_threads();
return;
}
mutex_enter(&(trx_doublewrite->mutex));
/* Write first to doublewrite buffer blocks. We use synchronous
aio and thus know that file write has been completed when the
control returns. */
if (trx_doublewrite->first_free == 0) {
mutex_exit(&(trx_doublewrite->mutex));
return;
}
for (i = 0; i < trx_doublewrite->first_free; i++) {
block = trx_doublewrite->buf_block_arr[i];
ut_a(block->state == BUF_BLOCK_FILE_PAGE);
if (mach_read_from_4(block->frame + FIL_PAGE_LSN + 4)
!= mach_read_from_4(block->frame + UNIV_PAGE_SIZE
- FIL_PAGE_END_LSN_OLD_CHKSUM + 4)) {
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: ERROR: The page to be written"
" seems corrupt!\n"
"InnoDB: The lsn fields do not match!"
" Noticed in the buffer pool\n"
"InnoDB: before posting to the"
" doublewrite buffer.\n");
}
if (block->check_index_page_at_flush
&& !page_simple_validate(block->frame)) {
buf_page_print(block->frame);
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: Apparent corruption of an"
" index page n:o %lu in space %lu\n"
"InnoDB: to be written to data file."
" We intentionally crash server\n"
"InnoDB: to prevent corrupt data"
" from ending up in data\n"
"InnoDB: files.\n",
(ulong) block->offset, (ulong) block->space);
ut_error;
}
}
/* increment the doublewrite flushed pages counter */
srv_dblwr_pages_written+= trx_doublewrite->first_free;
srv_dblwr_writes++;
if (trx_doublewrite->first_free > TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
len = TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * UNIV_PAGE_SIZE;
} else {
len = trx_doublewrite->first_free * UNIV_PAGE_SIZE;
}
fil_io(OS_FILE_WRITE,
TRUE, TRX_SYS_SPACE,
trx_doublewrite->block1, 0, len,
(void*)trx_doublewrite->write_buf, NULL);
write_buf = trx_doublewrite->write_buf;
for (len2 = 0; len2 + UNIV_PAGE_SIZE <= len; len2 += UNIV_PAGE_SIZE) {
if (mach_read_from_4(write_buf + len2 + FIL_PAGE_LSN + 4)
!= mach_read_from_4(write_buf + len2 + UNIV_PAGE_SIZE
- FIL_PAGE_END_LSN_OLD_CHKSUM + 4)) {
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: ERROR: The page to be written"
" seems corrupt!\n"
"InnoDB: The lsn fields do not match!"
" Noticed in the doublewrite block1.\n");
}
}
if (trx_doublewrite->first_free > TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
len = (trx_doublewrite->first_free
- TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) * UNIV_PAGE_SIZE;
fil_io(OS_FILE_WRITE,
TRUE, TRX_SYS_SPACE,
trx_doublewrite->block2, 0, len,
(void*)(trx_doublewrite->write_buf
+ TRX_SYS_DOUBLEWRITE_BLOCK_SIZE
* UNIV_PAGE_SIZE),
NULL);
write_buf = trx_doublewrite->write_buf
+ TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * UNIV_PAGE_SIZE;
for (len2 = 0; len2 + UNIV_PAGE_SIZE <= len;
len2 += UNIV_PAGE_SIZE) {
if (mach_read_from_4(write_buf + len2
+ FIL_PAGE_LSN + 4)
!= mach_read_from_4(write_buf + len2
+ UNIV_PAGE_SIZE
- FIL_PAGE_END_LSN_OLD_CHKSUM
+ 4)) {
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: ERROR: The page to be"
" written seems corrupt!\n"
"InnoDB: The lsn fields do not match!"
" Noticed in"
" the doublewrite block2.\n");
}
}
}
/* Now flush the doublewrite buffer data to disk */
fil_flush(TRX_SYS_SPACE);
/* We know that the writes have been flushed to disk now
and in recovery we will find them in the doublewrite buffer
blocks. Next do the writes to the intended positions. */
for (i = 0; i < trx_doublewrite->first_free; i++) {
block = trx_doublewrite->buf_block_arr[i];
if (mach_read_from_4(block->frame + FIL_PAGE_LSN + 4)
!= mach_read_from_4(block->frame + UNIV_PAGE_SIZE
- FIL_PAGE_END_LSN_OLD_CHKSUM + 4)) {
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: ERROR: The page to be written"
" seems corrupt!\n"
"InnoDB: The lsn fields do not match!"
" Noticed in the buffer pool\n"
"InnoDB: after posting and flushing"
" the doublewrite buffer.\n"
"InnoDB: Page buf fix count %lu,"
" io fix %lu, state %lu\n",
(ulong)block->buf_fix_count,
(ulong)block->io_fix,
(ulong)block->state);
}
ut_a(block->state == BUF_BLOCK_FILE_PAGE);
fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER,
FALSE, block->space, block->offset, 0, UNIV_PAGE_SIZE,
(void*)block->frame, (void*)block);
}
/* Wake possible simulated aio thread to actually post the
writes to the operating system */
os_aio_simulated_wake_handler_threads();
/* Wait that all async writes to tablespaces have been posted to
the OS */
os_aio_wait_until_no_pending_writes();
/* Now we flush the data to disk (for example, with fsync) */
fil_flush_file_spaces(FIL_TABLESPACE);
/* We can now reuse the doublewrite memory buffer: */
trx_doublewrite->first_free = 0;
mutex_exit(&(trx_doublewrite->mutex));
}
/************************************************************************
Posts a buffer page for writing. If the doublewrite memory buffer is
full, calls buf_flush_buffered_writes and waits for for free space to
appear. */
static
void
buf_flush_post_to_doublewrite_buf(
/*==============================*/
buf_block_t* block) /* in: buffer block to write */
{
try_again:
mutex_enter(&(trx_doublewrite->mutex));
ut_a(block->state == BUF_BLOCK_FILE_PAGE);
if (trx_doublewrite->first_free
>= 2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
mutex_exit(&(trx_doublewrite->mutex));
buf_flush_buffered_writes();
goto try_again;
}
ut_memcpy(trx_doublewrite->write_buf
+ UNIV_PAGE_SIZE * trx_doublewrite->first_free,
block->frame, UNIV_PAGE_SIZE);
trx_doublewrite->buf_block_arr[trx_doublewrite->first_free] = block;
trx_doublewrite->first_free++;
if (trx_doublewrite->first_free
>= 2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
mutex_exit(&(trx_doublewrite->mutex));
buf_flush_buffered_writes();
return;
}
mutex_exit(&(trx_doublewrite->mutex));
}
/************************************************************************
Initializes a page for writing to the tablespace. */
void
buf_flush_init_for_writing(
/*=======================*/
byte* page, /* in: page */
dulint newest_lsn, /* in: newest modification lsn to the page */
ulint space, /* in: space id */
ulint page_no) /* in: page number */
{
/* Write the newest modification lsn to the page header and trailer */
mach_write_to_8(page + FIL_PAGE_LSN, newest_lsn);
mach_write_to_8(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM,
newest_lsn);
/* Write the page number and the space id */
mach_write_to_4(page + FIL_PAGE_OFFSET, page_no);
mach_write_to_4(page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID, space);
/* Store the new formula checksum */
mach_write_to_4(page + FIL_PAGE_SPACE_OR_CHKSUM,
srv_use_checksums
? buf_calc_page_new_checksum(page)
: BUF_NO_CHECKSUM_MAGIC);
/* We overwrite the first 4 bytes of the end lsn field to store
the old formula checksum. Since it depends also on the field
FIL_PAGE_SPACE_OR_CHKSUM, it has to be calculated after storing the
new formula checksum. */
mach_write_to_4(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM,
srv_use_checksums
? buf_calc_page_old_checksum(page)
: BUF_NO_CHECKSUM_MAGIC);
}
/************************************************************************
Does an asynchronous write of a buffer page. NOTE: in simulated aio and
also when the doublewrite buffer is used, we must call
buf_flush_buffered_writes after we have posted a batch of writes! */
static
void
buf_flush_write_block_low(
/*======================*/
buf_block_t* block) /* in: buffer block to write */
{
#ifdef UNIV_LOG_DEBUG
static ibool univ_log_debug_warned;
#endif /* UNIV_LOG_DEBUG */
ut_a(block->state == BUF_BLOCK_FILE_PAGE);
#ifdef UNIV_IBUF_DEBUG
ut_a(ibuf_count_get(block->space, block->offset) == 0);
#endif
ut_ad(!ut_dulint_is_zero(block->newest_modification));
#ifdef UNIV_LOG_DEBUG
if (!univ_log_debug_warned) {
univ_log_debug_warned = TRUE;
fputs("Warning: cannot force log to disk if"
" UNIV_LOG_DEBUG is defined!\n"
"Crash recovery will not work!\n",
stderr);
}
#else
/* Force the log to the disk before writing the modified block */
log_write_up_to(block->newest_modification, LOG_WAIT_ALL_GROUPS, TRUE);
#endif
buf_flush_init_for_writing(block->frame, block->newest_modification,
block->space, block->offset);
if (!srv_use_doublewrite_buf || !trx_doublewrite) {
fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER,
FALSE, block->space, block->offset, 0, UNIV_PAGE_SIZE,
(void*)block->frame, (void*)block);
} else {
buf_flush_post_to_doublewrite_buf(block);
}
}
/************************************************************************
Writes a page asynchronously from the buffer buf_pool to a file, if it can be
found in the buf_pool and it is in a flushable state. NOTE: in simulated aio
we must call os_aio_simulated_wake_handler_threads after we have posted a batch
of writes! */
static
ulint
buf_flush_try_page(
/*===============*/
/* out: 1 if a page was flushed, 0 otherwise */
ulint space, /* in: space id */
ulint offset, /* in: page offset */
ulint flush_type) /* in: BUF_FLUSH_LRU, BUF_FLUSH_LIST, or
BUF_FLUSH_SINGLE_PAGE */
{
buf_block_t* block;
ibool locked;
ut_ad(flush_type == BUF_FLUSH_LRU || flush_type == BUF_FLUSH_LIST
|| flush_type == BUF_FLUSH_SINGLE_PAGE);
mutex_enter(&(buf_pool->mutex));
block = buf_page_hash_get(space, offset);
ut_a(!block || block->state == BUF_BLOCK_FILE_PAGE);
if (!block) {
mutex_exit(&(buf_pool->mutex));
return(0);
}
mutex_enter(&block->mutex);
if (flush_type == BUF_FLUSH_LIST
&& buf_flush_ready_for_flush(block, flush_type)) {
block->io_fix = BUF_IO_WRITE;
/* If AWE is enabled and the page is not mapped to a frame,
then map it */
if (block->frame == NULL) {
ut_a(srv_use_awe);
/* We set second parameter TRUE because the block is
in the LRU list and we must put it to
awe_LRU_free_mapped list once mapped to a frame */
buf_awe_map_page_to_frame(block, TRUE);
}
block->flush_type = flush_type;
if (buf_pool->n_flush[flush_type] == 0) {
os_event_reset(buf_pool->no_flush[flush_type]);
}
(buf_pool->n_flush[flush_type])++;
locked = FALSE;
/* If the simulated aio thread is not running, we must
not wait for any latch, as we may end up in a deadlock:
if buf_fix_count == 0, then we know we need not wait */
if (block->buf_fix_count == 0) {
rw_lock_s_lock_gen(&(block->lock), BUF_IO_WRITE);
locked = TRUE;
}
mutex_exit(&block->mutex);
mutex_exit(&(buf_pool->mutex));
if (!locked) {
buf_flush_buffered_writes();
rw_lock_s_lock_gen(&(block->lock), BUF_IO_WRITE);
}
#ifdef UNIV_DEBUG
if (buf_debug_prints) {
fprintf(stderr,
"Flushing page space %lu, page no %lu \n",
(ulong) block->space, (ulong) block->offset);
}
#endif /* UNIV_DEBUG */
buf_flush_write_block_low(block);
return(1);
} else if (flush_type == BUF_FLUSH_LRU
&& buf_flush_ready_for_flush(block, flush_type)) {
/* VERY IMPORTANT:
Because any thread may call the LRU flush, even when owning
locks on pages, to avoid deadlocks, we must make sure that the
s-lock is acquired on the page without waiting: this is
accomplished because in the if-condition above we require
the page not to be bufferfixed (in function
..._ready_for_flush). */
block->io_fix = BUF_IO_WRITE;
/* If AWE is enabled and the page is not mapped to a frame,
then map it */
if (block->frame == NULL) {
ut_a(srv_use_awe);
/* We set second parameter TRUE because the block is
in the LRU list and we must put it to
awe_LRU_free_mapped list once mapped to a frame */
buf_awe_map_page_to_frame(block, TRUE);
}
block->flush_type = flush_type;
if (buf_pool->n_flush[flush_type] == 0) {
os_event_reset(buf_pool->no_flush[flush_type]);
}
(buf_pool->n_flush[flush_type])++;
rw_lock_s_lock_gen(&(block->lock), BUF_IO_WRITE);
/* Note that the s-latch is acquired before releasing the
buf_pool mutex: this ensures that the latch is acquired
immediately. */
mutex_exit(&block->mutex);
mutex_exit(&(buf_pool->mutex));
buf_flush_write_block_low(block);
return(1);
} else if (flush_type == BUF_FLUSH_SINGLE_PAGE
&& buf_flush_ready_for_flush(block, flush_type)) {
block->io_fix = BUF_IO_WRITE;
/* If AWE is enabled and the page is not mapped to a frame,
then map it */
if (block->frame == NULL) {
ut_a(srv_use_awe);
/* We set second parameter TRUE because the block is
in the LRU list and we must put it to
awe_LRU_free_mapped list once mapped to a frame */
buf_awe_map_page_to_frame(block, TRUE);
}
block->flush_type = flush_type;
if (buf_pool->n_flush[block->flush_type] == 0) {
os_event_reset(buf_pool->no_flush[block->flush_type]);
}
(buf_pool->n_flush[flush_type])++;
mutex_exit(&block->mutex);
mutex_exit(&(buf_pool->mutex));
rw_lock_s_lock_gen(&(block->lock), BUF_IO_WRITE);
#ifdef UNIV_DEBUG
if (buf_debug_prints) {
fprintf(stderr,
"Flushing single page space %lu,"
" page no %lu \n",
(ulong) block->space,
(ulong) block->offset);
}
#endif /* UNIV_DEBUG */
buf_flush_write_block_low(block);
return(1);
}
mutex_exit(&block->mutex);
mutex_exit(&(buf_pool->mutex));
return(0);
}
/***************************************************************
Flushes to disk all flushable pages within the flush area. */
static
ulint
buf_flush_try_neighbors(
/*====================*/
/* out: number of pages flushed */
ulint space, /* in: space id */
ulint offset, /* in: page offset */
ulint flush_type) /* in: BUF_FLUSH_LRU or BUF_FLUSH_LIST */
{
buf_block_t* block;
ulint low, high;
ulint count = 0;
ulint i;
ut_ad(flush_type == BUF_FLUSH_LRU || flush_type == BUF_FLUSH_LIST);
low = (offset / BUF_FLUSH_AREA) * BUF_FLUSH_AREA;
high = (offset / BUF_FLUSH_AREA + 1) * BUF_FLUSH_AREA;
if (UT_LIST_GET_LEN(buf_pool->LRU) < BUF_LRU_OLD_MIN_LEN) {
/* If there is little space, it is better not to flush any
block except from the end of the LRU list */
low = offset;
high = offset + 1;
}
/* fprintf(stderr, "Flush area: low %lu high %lu\n", low, high); */
if (high > fil_space_get_size(space)) {
high = fil_space_get_size(space);
}
mutex_enter(&(buf_pool->mutex));
for (i = low; i < high; i++) {
block = buf_page_hash_get(space, i);
ut_a(!block || block->state == BUF_BLOCK_FILE_PAGE);
if (!block) {
continue;
} else if (flush_type == BUF_FLUSH_LRU && i != offset
&& !block->old) {
/* We avoid flushing 'non-old' blocks in an LRU flush,
because the flushed blocks are soon freed */
continue;
} else {
mutex_enter(&block->mutex);
if (buf_flush_ready_for_flush(block, flush_type)
&& (i == offset || block->buf_fix_count == 0)) {
/* We only try to flush those
neighbors != offset where the buf fix count is
zero, as we then know that we probably can
latch the page without a semaphore wait.
Semaphore waits are expensive because we must
flush the doublewrite buffer before we start
waiting. */
mutex_exit(&block->mutex);
mutex_exit(&(buf_pool->mutex));
/* Note: as we release the buf_pool mutex
above, in buf_flush_try_page we cannot be sure
the page is still in a flushable state:
therefore we check it again inside that
function. */
count += buf_flush_try_page(space, i,
flush_type);
mutex_enter(&(buf_pool->mutex));
} else {
mutex_exit(&block->mutex);
}
}
}
mutex_exit(&(buf_pool->mutex));
return(count);
}
/***********************************************************************
This utility flushes dirty blocks from the end of the LRU list or flush_list.
NOTE 1: in the case of an LRU flush the calling thread may own latches to
pages: to avoid deadlocks, this function must be written so that it cannot
end up waiting for these latches! NOTE 2: in the case of a flush list flush,
the calling thread is not allowed to own any latches on pages! */
ulint
buf_flush_batch(
/*============*/
/* out: number of blocks for which the write
request was queued; ULINT_UNDEFINED if there
was a flush of the same type already running */
ulint flush_type, /* in: BUF_FLUSH_LRU or BUF_FLUSH_LIST; if
BUF_FLUSH_LIST, then the caller must not own
any latches on pages */
ulint min_n, /* in: wished minimum mumber of blocks flushed
(it is not guaranteed that the actual number
is that big, though) */
dulint lsn_limit) /* in the case BUF_FLUSH_LIST all blocks whose
oldest_modification is smaller than this
should be flushed (if their number does not
exceed min_n), otherwise ignored */
{
buf_block_t* block;
ulint page_count = 0;
ulint old_page_count;
ulint space;
ulint offset;
ibool found;
ut_ad((flush_type == BUF_FLUSH_LRU)
|| (flush_type == BUF_FLUSH_LIST));
#ifdef UNIV_SYNC_DEBUG
ut_ad((flush_type != BUF_FLUSH_LIST)
|| sync_thread_levels_empty_gen(TRUE));
#endif /* UNIV_SYNC_DEBUG */
mutex_enter(&(buf_pool->mutex));
if ((buf_pool->n_flush[flush_type] > 0)
|| (buf_pool->init_flush[flush_type] == TRUE)) {
/* There is already a flush batch of the same type running */
mutex_exit(&(buf_pool->mutex));
return(ULINT_UNDEFINED);
}
(buf_pool->init_flush)[flush_type] = TRUE;
for (;;) {
/* If we have flushed enough, leave the loop */
if (page_count >= min_n) {
break;
}
/* Start from the end of the list looking for a suitable
block to be flushed. */
if (flush_type == BUF_FLUSH_LRU) {
block = UT_LIST_GET_LAST(buf_pool->LRU);
} else {
ut_ad(flush_type == BUF_FLUSH_LIST);
block = UT_LIST_GET_LAST(buf_pool->flush_list);
if (!block
|| (ut_dulint_cmp(block->oldest_modification,
lsn_limit) >= 0)) {
/* We have flushed enough */
break;
}
}
found = FALSE;
/* Note that after finding a single flushable page, we try to
flush also all its neighbors, and after that start from the
END of the LRU list or flush list again: the list may change
during the flushing and we cannot safely preserve within this
function a pointer to a block in the list! */
while ((block != NULL) && !found) {
ut_a(block->state == BUF_BLOCK_FILE_PAGE);
mutex_enter(&block->mutex);
if (buf_flush_ready_for_flush(block, flush_type)) {
found = TRUE;
space = block->space;
offset = block->offset;
mutex_exit(&block->mutex);
mutex_exit(&(buf_pool->mutex));
old_page_count = page_count;
/* Try to flush also all the neighbors */
page_count += buf_flush_try_neighbors(
space, offset, flush_type);
/* fprintf(stderr,
"Flush type %lu, page no %lu, neighb %lu\n",
flush_type, offset,
page_count - old_page_count); */
mutex_enter(&(buf_pool->mutex));
} else if (flush_type == BUF_FLUSH_LRU) {
mutex_exit(&block->mutex);
block = UT_LIST_GET_PREV(LRU, block);
} else {
ut_ad(flush_type == BUF_FLUSH_LIST);
mutex_exit(&block->mutex);
block = UT_LIST_GET_PREV(flush_list, block);
}
}
/* If we could not find anything to flush, leave the loop */
if (!found) {
break;
}
}
(buf_pool->init_flush)[flush_type] = FALSE;
if ((buf_pool->n_flush[flush_type] == 0)
&& (buf_pool->init_flush[flush_type] == FALSE)) {
/* The running flush batch has ended */
os_event_set(buf_pool->no_flush[flush_type]);
}
mutex_exit(&(buf_pool->mutex));
buf_flush_buffered_writes();
#ifdef UNIV_DEBUG
if (buf_debug_prints && page_count > 0) {
ut_a(flush_type == BUF_FLUSH_LRU
|| flush_type == BUF_FLUSH_LIST);
fprintf(stderr, flush_type == BUF_FLUSH_LRU
? "Flushed %lu pages in LRU flush\n"
: "Flushed %lu pages in flush list flush\n",
(ulong) page_count);
}
#endif /* UNIV_DEBUG */
srv_buf_pool_flushed += page_count;
return(page_count);
}
/**********************************************************************
Waits until a flush batch of the given type ends */
void
buf_flush_wait_batch_end(
/*=====================*/
ulint type) /* in: BUF_FLUSH_LRU or BUF_FLUSH_LIST */
{
ut_ad((type == BUF_FLUSH_LRU) || (type == BUF_FLUSH_LIST));
os_event_wait(buf_pool->no_flush[type]);
}
/**********************************************************************
Gives a recommendation of how many blocks should be flushed to establish
a big enough margin of replaceable blocks near the end of the LRU list
and in the free list. */
static
ulint
buf_flush_LRU_recommendation(void)
/*==============================*/
/* out: number of blocks which should be flushed
from the end of the LRU list */
{
buf_block_t* block;
ulint n_replaceable;
ulint distance = 0;
mutex_enter(&(buf_pool->mutex));
n_replaceable = UT_LIST_GET_LEN(buf_pool->free);
block = UT_LIST_GET_LAST(buf_pool->LRU);
while ((block != NULL)
&& (n_replaceable < BUF_FLUSH_FREE_BLOCK_MARGIN
+ BUF_FLUSH_EXTRA_MARGIN)
&& (distance < BUF_LRU_FREE_SEARCH_LEN)) {
mutex_enter(&block->mutex);
if (buf_flush_ready_for_replace(block)) {
n_replaceable++;
}
mutex_exit(&block->mutex);
distance++;
block = UT_LIST_GET_PREV(LRU, block);
}
mutex_exit(&(buf_pool->mutex));
if (n_replaceable >= BUF_FLUSH_FREE_BLOCK_MARGIN) {
return(0);
}
return(BUF_FLUSH_FREE_BLOCK_MARGIN + BUF_FLUSH_EXTRA_MARGIN
- n_replaceable);
}
/*************************************************************************
Flushes pages from the end of the LRU list if there is too small a margin
of replaceable pages there or in the free list. VERY IMPORTANT: this function
is called also by threads which have locks on pages. To avoid deadlocks, we
flush only pages such that the s-lock required for flushing can be acquired
immediately, without waiting. */
void
buf_flush_free_margin(void)
/*=======================*/
{
ulint n_to_flush;
ulint n_flushed;
n_to_flush = buf_flush_LRU_recommendation();
if (n_to_flush > 0) {
n_flushed = buf_flush_batch(BUF_FLUSH_LRU, n_to_flush,
ut_dulint_zero);
if (n_flushed == ULINT_UNDEFINED) {
/* There was an LRU type flush batch already running;
let us wait for it to end */
buf_flush_wait_batch_end(BUF_FLUSH_LRU);
}
}
}
/**********************************************************************
Validates the flush list. */
static
ibool
buf_flush_validate_low(void)
/*========================*/
/* out: TRUE if ok */
{
buf_block_t* block;
dulint om;
UT_LIST_VALIDATE(flush_list, buf_block_t, buf_pool->flush_list);
block = UT_LIST_GET_FIRST(buf_pool->flush_list);
while (block != NULL) {
om = block->oldest_modification;
ut_a(block->state == BUF_BLOCK_FILE_PAGE);
ut_a(ut_dulint_cmp(om, ut_dulint_zero) > 0);
block = UT_LIST_GET_NEXT(flush_list, block);
if (block) {
ut_a(ut_dulint_cmp(om, block->oldest_modification)
>= 0);
}
}
return(TRUE);
}
/**********************************************************************
Validates the flush list. */
ibool
buf_flush_validate(void)
/*====================*/
/* out: TRUE if ok */
{
ibool ret;
mutex_enter(&(buf_pool->mutex));
ret = buf_flush_validate_low();
mutex_exit(&(buf_pool->mutex));
return(ret);
}