ORACLE数据文件和控制文件头部.doc

为了回答关于深入浅出Oracle中的一些疑问，引出本系列文章，讨论链接参考:/609499.html在上一讲中，我们说过：当我们使用file_hdrs事件来转储数据文件头信息时，Oracle会转储两部分信息，一部分来自控制文件，一部分来自数据文件，在数据库启动过程中，这两部分信息要用来进行启动验证。在数据库open的过程中,Oracle要进行检查中包含以下两个过程:第一次检查数据文件头中的Checkpoint cnt是否与对应控制文件中的Checkpoint cnt一致.如果相等,进行第二次检查.第二次检查数据文件头的开始SCN和对应控制文件中的结束SCN是否一致如果结束SCN等于开始SCN,则不需要对那个文件进行恢复.对每个数据文件都完成检查后,打开数据库.同时将每个数据文件的结束SCN设置为无穷大.通过以下过程我们来进一步说明一下这个内容。我们来看以下来自控制文件部分(选取一个文件测试): DATA FILE #4: (name #4) /opt/oracle/oradata/eygle/eygle01.dbfcreation size=0 block size=8192 status=0xe head=4 tail=4 dup=1tablespace 4, index=4 krfil=4 prev_file=0unrecoverable scn: 0x0000.00000000 01/01/1988 00:00:00Checkpoint cnt:58 scn: 0x0000.002ac8ee 08/11/2006 09:48:29Stop scn: 0x0000.002ac8ee 08/11/2006 09:48:29Creation Checkpointed at scn: 0x0000.0015078d 06/06/2006 09:41:54thread:0 rba:(0x0.0.0).aux_file is NOT DEFINED 这部分中包含的重要信息有:检查点计数: Checkpoint cnt:58检查点SCN: scn: 0x0000.002ac8ee 08/11/2006 09:48:29数据文件Stop SCN:Stop scn: 0x0000.002ac8ee 08/11/2006 09:48:29我们再看来自数据文件头的信息:FILE HEADER:Software vsn=153092096=0x9200000, Compatibility Vsn=134217728=0x8000000Db ID=1407686520=0x53e79778, Db Name=EYGLEActivation ID=0=0x0Control Seq=979=0x3d3, File size=1280=0x500File Number=4, Blksiz=8192, File Type=3 DATATablespace #4 - EYGLE rel_fn:4 Creation at scn: 0x0000.0015078d 06/06/2006 09:41:54Backup taken at scn: 0x0000.00000000 01/01/1988 00:00:00 thread:0reset logs count:0x232bee1f scn: 0x0000.0007c781 recovered at 08/10/2006 20:57:53status:0x0 root dba:0x00000000 chkpt cnt: 58 ctl cnt:57begin-hot-backup file size: 0Checkpointed at scn: 0x0000.002ac8ee 08/11/2006 09:48:29.这部分中包含的重要信息有:检查点SCN: Checkpointed at scn: 0x0000.002ac8ee 08/11/2006 09:48:29检查点计数: chkpt cnt: 58 ctl cnt:57这两者都和控制文件中所记录的一致。如果这两者一致，数据库启动时就能通过验证，启动数据库。那么如果不一致呢？Oracle则请求进行恢复。我们看，从备份中恢复eygle01.dbf文件.首先第一部分从控制文件中获得的信息是相同的: DATA FILE #4: (name #4) /opt/oracle/oradata/eygle/eygle01.dbfcreation size=0 block size=8192 status=0xe head=4 tail=4 dup=1tablespace 4, index=4 krfil=4 prev_file=0unrecoverable scn: 0x0000.00000000 01/01/1988 00:00:00Checkpoint cnt:58 scn: 0x0000.002ac8ee 08/11/2006 09:48:29Stop scn: 0x0000.002ac8ee 08/11/2006 09:48:29Creation Checkpointed at scn: 0x0000.0015078d 06/06/2006 09:41:54.aux_file is NOT DEFINED 检查点计数: Checkpoint cnt:58检查点SCN: scn: 0x0000.002ac8ee 08/11/2006 09:48:29数据文件Stop SCN:Stop scn: 0x0000.002ac8ee 08/11/2006 09:48:29而从文件头中获得的备份文件信息则是： FILE HEADER:Software vsn=153092096=0x9200000, Compatibility Vsn=134217728=0x8000000Db ID=1407686520=0x53e79778, Db Name=EYGLEActivation ID=0=0x0Control Seq=973=0x3cd, File size=1280=0x500File Number=4, Blksiz=8192, File Type=3 DATATablespace #4 - EYGLE rel_fn:4 Creation at scn: 0x0000.0015078d 06/06/2006 09:41:54Backup taken at scn: 0x0000.00000000 01/01/1988 00:00:00 thread:0reset logs count:0x232bee1f scn: 0x0000.0007c781 recovered at 08/10/2006 20:57:53status:0x0 root dba:0x00000000 chkpt cnt: 53 ctl cnt:52begin-hot-backup file size: 0Checkpointed at scn: 0x0000.002ac5f9 08/10/2006 20:58:21.我们看到此时备份文件的信息：检查点是：Checkpointed at scn: 0x0000.002ac5f9 08/10/2006 20:58:21检查点计数为:chkpt cnt: 53 ctl cnt:52这两者不再一致，首先是检查点技术不一致，当前文件的chkpt cnt为53，小于控制文件中记录的58，Oracle可以判断文件是从备份中恢复的，或者文件故障，需要进行介质恢复。我们看如果此时我们试图打开数据库，则Oracle提示文件需要介质恢复: SQL alter database open;alter database open*ERROR at line 1:ORA-01113: file 4 needs media recoveryORA-01110: data file 4: /opt/oracle/oradata/eygle/eygle01.dbf执行恢复: SQL recover datafile 4;Media recovery complete.我们看看恢复完成之后，控制文件和数据文件的变化.首先看控制文件的变化: DATA FILE #4: (name #4) /opt/oracle/oradata/eygle/eygle01.dbfcreation size=0 block size=8192 status=0xe head=4 tail=4 dup=1tablespace 4, index=4 krfil=4 prev_file=0unrecoverable scn: 0x0000.00000000 01/01/1988 00:00:00Checkpoint cnt:59 scn: 0x0000.002ac8ee 08/11/2006 09:48:29Stop scn: 0x0000.002ac8ed 08/11/2006 09:48:29Creation Checkpointed at scn: 0x0000.0015078d 06/06/2006 09:41:54.检查点计数: Checkpoint cnt:59执行了恢复之后，检查点计数较前增加了1检查点SCN: scn: 0x0000.002ac8ee 08/11/2006 09:48:29数据文件Stop scn: 0x0000.002ac8ed 08/11/2006 09:48:29数据文件Stop scn和数据文件进行了同步。数据文件头信息: FILE HEADER:Software vsn=153092096=0x9200000, Compatibility Vsn=134217728=0x8000000Db ID=1407686520=0x53e79778, Db Name=EYGLEActivation ID=0=0x0Control Seq=983=0x3d7, File size=1280=0x500File Number=4, Blksiz=8192, File Type=3 DATATablespace #4 - EYGLE rel_fn:4 Creation at scn: 0x0000.0015078d 06/06/2006 09:41:54Backup taken at scn: 0x0000.00000000 01/01/1988 00:00:00 thread:0reset logs count:0x232bee1f scn: 0x0000.0007c781 recovered at 08/11/2006 10:11:26status:0x0 root dba:0x00000000 chkpt cnt: 59 ctl cnt:58begin-hot-backup file size: 0Checkpointed at scn: 0x0000.002ac8ed 08/11/2006 09:48:29.我们看到此时数据文件的信息：检查点是：Checkpointed at scn: 0x0000.002ac8ed 08/11/2006 09:48:29这个检查点和控制文件中记录的stop scn一致，数据库启动可以顺利进行。检查点计数为:chkpt cnt: 59 ctl cnt:58我们打开数据库: SQL alter database open;Database altered.SQL alter session set events immediate trace name file_hdrs level 10;Session altered.此时数据库恢复正常运行。控制文件信息如下： DATA FILE #4: (name #4) /opt/oracle/oradata/eygle/eygle01.dbfcreation size=0 block size=8192 status=0xe head=4 tail=4 dup=1tablespace 4, index=4 krfil=4 prev_file=0unrecoverable scn: 0x0000.00000000 01/01/1988 00:00:00Checkpoint cnt:60 scn: 0x0000.002ac8ef 08/11/2006 10:19:30Stop scn: 0xffff.ffffffff 08/11/2006 09:48:29Creation Checkpointed at scn: 0x0000.0015078d 06/06/2006 09:41:54此时stop scn被置为无穷大。数据文件头信息如下:FILE HEADER:Software vsn=153092096=0x9200000, Compatibility Vsn=134217728=0x8000000Db ID=1407686520=0x53e79778, Db Name=EYGLEActivation ID=0=0x0Control Seq=984=0x3d8, File size=1280=0x500File Number=4, Blksiz=8192, File Type=3 DATATablespace #4 - EYGLE rel_fn:4 Creation at scn: 0x0000.0015078d 06/06/2006 09:41:54Backup taken at scn: 0x0000.00000000 01/01/1988 00:00:00 thread:0reset logs count:0x232bee1f scn: 0x0000.0007c781 recovered at 08/11/2006 10:11:26status:0x4 root dba:0x00000000 chkpt cnt: 60 ctl cnt:59begin-hot-backup file size: 0Checkpointed at scn: 0x0000.002ac8ef 08/11/2006 10:19:30未完待续.datafile block block size :8192Offset 0 1 2 3 4 5 6 7 8 9 a b c d e f 00014000 06 A2 00 00 0A 00 40 01 0E 89 43 00 00 00 05 02 type frmt spare1/2_kcbh rdba scn seq flg 1 : 20 bytestype: 0x06=trans data defined in kcb.h frmt: 8i9i 都是0x02 10.1.0 2k: 0x62 4k:0x82 8k:0xa2 16k:0xc2 (logfile 0x22 512 bytes) spare1/2_kcbh: ub1 spare1_kcbh this field is no longer used (old inc#, now always 0) ub1 spare2_kcbh this field is no longer used (old ts#, now always 0) rdba: 0x0140000a 转换成2进制后它的前10 bit 表示file id 后22 bit 表示的block id 可以看出一个tablespace 可以有1023 个datafile ,每个datafile可以有4M 的block10G 出现的 big datafile 这里表示的就是block id了 没有file id 9.2.0试验过一个tablespace可以有1023个datafile 一个object可以存放在1023个datafile中 scn: scn: 0x0000.0043890e seq: A sequence number incremented for each change to a block at the same SCN A new SCN is allocated if the sequence number wraps. 同一个SCN影响这个block中的行数大于 254 行就会为这个事务分配一个新的SCN 如下面的操作就可能引起同一个SCN但影响的同一个block 中的行超过254行 delete from table_name 影响的行数(最大254) 是用从 0x01 到 0xfe 表示的当这个byte 的数据为 0xff 的时候标志这个 block 坏调了- ora-01578Sequence number:SEQ - 0 /* non-logged changes - do not advance seq# */SEQ - (UB1MAXVAL-1)/* maximum possible sequence number */SEQ - (UB1MAXVAL) /* seq# to indicate a block is corrupt,equal to FF. soft corrupt*/ 0xff : When present it indicates that the block has been marked as corrupt by Oracle. either by the db_block_checking functionality or the equivalent events (10210 for data blocks, 10211 for index blocks, and 10212 for cluster blocks) when making a database change, or by the DBMS_REPAIR.FIX_CORRUPT_BLOCKS procedure, or by PMON after an unsuccessful online block recovery attempt while recovering a failed process, or by RMAN during a BACKUP, COPY or VALIDATE command with the CHECK LOGICAL option. Logical corruptions are normally due to either recovery through a NOLOGGING operation, or an Oracle software bug. flg: as defined in kcbh.h #define KCBHFNEW 0x01 /* new block - zeroed data area */#define KCBHFDLC 0x02 /* Delayed Logging Change advance SCN/seq */#define KCBHFCKV 0x04 /* ChecK Value saved-block xors to zero */#define KCBHFTMP 0x08 /* Temporary block */这是一个可以组合的值 也就是说有为 6 的时候是 2,4 两种情况的组合Block structure as defined in kcbh.h:struct kcbhub1 type_kcbh; /* Block type* / ub1 frmt_kcbh; /* #define KCBH_FRMT8 2 */ ub1 spare1_kcbh;ub1 spare2_kcbh;krdba rdba_kcbh; /* relative DBA /ub4 bas_kcbh; /* base of SCN */ub2 wrp_kcbh; /* wrap of SCN */ub1 seq_kcbh; /* sequence # of changes at same scn */ub1 flg_kcbh;ub2 chkval_kcbh; 00014010 00 00 00 00 01 00 17 00 54 D2 00 00 0A 89 43 00 chkval spare3_kcbh typ ? seg/obj csc spare3_kcbh : ub2 spare3_kcbh 2 : 24 bytes (总计44bytes)typ : 1 - DATA 2 index 改成3了在10.1.0 上引起了ora-6002032然后ORA-27101: shared memory realm does not existoracle进行查询的时候是根据 obj$表中的情况来判断对象的类型的,不是根据这个typ也就是说如果有一个表但改变表中block的这个标志位，一样可以查询出数据来,但dump block 时会出错,ORA-00600: 内部错误代码，自变量: 4555, 0, , , , , , 错误中的 0 就是typ对应的数据在10G中改变它后update这个block的数据commit可以但rollback的报错 ? 见过有其他值 但用编辑器改这个值 在 dump 文件中显示不出来变化 seg/obj: 0xd254 csc : 0x00.43890a The SCN at which the last full cleanout was performed on the block 00014020 00 00 E8 1F 02 00 03 00 00 00 00 00 04 00 0C 00 csc ? itc ? flg fsl fnx xid 3 : 24 bytes * itl (2个itl总计92bytes)? 见过有其他值 但用编辑器改这个值 在 dump 文件中显示不出来变化 itc ITL 条目的个数 max 255超过会报ORA-02207 ORA-00060 ORA-00054 可能是没空间分配itl条目了或它的争用引起的在8i中 INITRANS default为1 , 9.2.0中 INITRANS default为2 flg indicates that the block is on a freelist. Otherwise the flag is -9i 的ASSM 的情况下这个值为 Eixora 上说他占用 2 bytes 但我下面的试验和他的结果有一定的出入我观察到的情况是 : Object id on Block? Y flg: O ver: 0x01 上面的3项是用同一个 byte 来表示的flg: O ver: 0x01 Object id on Block? Y 从我的观察中 dump 出来的文件中 flg ver Object id on Block他们共同占用的这个一个字节 他的规律可以从下面的情况看出2进制数据 flg ver Object id on Block?0x00 - 0x00 N0x01 0 0x00 N0x02 - 0x01 Y0x03 0 0x01 Y0x04 - 0x02 Y0x05 0 0x02 Y0x06 - 0x03 Y0x07 0 0x03 Y0x08 - 0x04 N0x09 0 0x04 N0x0a - 0x05 Y0x0b 0 0x05 Y0x0c - 0x06 Y0x0d 0 0x06 Y0x0e - 0x07 Y0x0f 0 0x07 Y0x10 . 类似上面的循环了 这种情况在9i上已经改变因为ASSM的出现 fsl : Index to the first slot on the ITL freelist. ITL TX freelist slot fnx : 自由列表中下一块的地址 Null if this block is not on a freelist 有数据例如: fnx: 0x1000029 00014030 50 18 00 00 96 14 80 00 B9 07 01 00 01 20 00 00 xid uba Lck Flag Scn/Fsc xid : Transaction ID (UndoSeg.Slot.Wrap)值可以用select XIDUSN, XIDSLOT,XIDSQN from v$transaction;查到 This is comprised of the rollback segment number (2 bytes), the slot numberin the transaction table of that rollback segment (2 bytes), and the numberof times use of that transaction table has wrapped (4 bytes). uba : Undo address (UndoDBA.SeqNo.RecordNo)The location of the undo for the most recent change to this block by this transaction. This is comprised of the DBA of the rollback segment block (4 bytes), the sequence number (2 bytes), and the record number for the change in that undo block (1 byte), plus 1 unused byte. Lck Flag: Lck 锁定的row数 这里还用到了下一个 byte 的数据2 对应的二进制表示为 0010 正好和dump文件中的 -U- 吻合flag 1 nibble C = Committed; U = Commit Upper Bound; T = Active at CSC; B = Rollback of this UBA gives before image of the ITL.- = transaction is active, or committed pending cleanoutC- = transaction has been committed and locks cleaned out-B- = this undo record contains the undo for this ITL entry-U- = transaction committed (maybe long ago); SCN is an upper bound-T = transaction was still active at block cleanout SCNLck 3 nibbles The number of row-level locks held in the block by this transaction. Scn/Fsc : If the transaction has been cleaned out, this is the commit SCN or an upper bound thereof. Otherwise the leading two bytes contain the free space credit for the transaction - that is, the number of bytes freed in the block by the transaction Scn = SCN of commited TX; Fsc = Free space credit (bytes) 00014040 0E 89 43 00 00 00 00 00 00 00 00 00 00 00 00 00 Scn/Fsc 第2条itl 这里没使用 00014050 00 00 00 00 00 00 00 00 00 00 00 00 00 01 01 00 第2条itl 这里没使用 flag ntab nrow 4 : 14 bytes 从这个flag位置开始是data区 也是下面的行的offset的起始地址flag : N=pctfree hit(clusters), F=dont put on free list K=flushable cluster keys. 当然还有别的标记: A . ntab : 这block中有几个table的数据 cluster这个就可能大于1 nrow : block 有多少行数据 00014060 FF FF 14 00 9B 1F 83 1F 83 1F 00 00 01 00 9B 1F frre fsbo fseo avsp tosp offs nrow row offs frre : First free row index entry. -1=you have to add one. fsbo : Free Space Begin offset 出去row dict 后面的可以放数据的空间的起始位置也可以看成是从这个区域的开始flag到最后一个 row offs占用的空间 fseo : Free Space End offset ( 9.2.0 )参与db_block_checking的计算剩余空间select 的时候oracle不是简单的根据offset定位row.这个值也是参与了定位row的 avsp : Available space in the block (pctfree and pctused) ORA-01578 tosp : Total available space when all TXs commit ( 9.2.0 )参与db_block_checking offs : 偏移量 用 cluster 的时候可以看出值 nrow : 这个table有多少行数据 row offs : 这行数据相对的起始位置 after delete & commit is 0xffff 00015FF0 00 00 00 00 00 00 00 2C 01 01 01 61 05 06 0E 89 fb lb cc length data block tail 5 : 用户数据6 : 4 bytes block tailfb : K = Cluster Key (Flags may change meaning if this is set to show HASH cluster) C = Cluster table member H = Head piece of row D = Deleted row F = First data piece L = Last data piece P = First column continues from previous piece N = Last column continues in next piece lb : 和上面的 ITL 的lck相对应 表示这行是否被 lock 了 cc : 有几列数据 这里只能表示255列 超过了就会有链接行 length : 这列的数据的长度是多少 0xfa ( 250 bytes ) 其实0xfb,0xfc,0xfd 也同样是250bytes 0xfe fb 00 ( 0xfb 00 表示的251 bytes 0xfe表示row的长度超过了250 bytes) 0xff 表示number 的 null 这也是oracle中null的表现形式排序的时候null最大了字段的数据超过250字节是就用3bytes来表示字段的长度,因为如果是long类型它的字段再长它在这个block中的数据的长度不会超过64K 所以最长用3bytes来表示行的长度已经够了.再长就链接行了 data : a block tail : 改这 block 最后的4 bytes 数据中的任意肯定ora-1578 第 1 byte : 对应开始的 seq 第 2 byte : 对应开始的 type 第3,4byte : 对应开始的scn的末2为 control file 这里是control seq 10.1.0lgoneONE.LG.OK create table a(v varchar2(4000) TABLESPACE t;Table created.10.1.0lgoneONE.LG.OK insert into a values(a);1 row created.Start dump data blocks tsn: 17 file#: 5 minblk 10 maxblk 10buffer tsn: 17 rdba: 0x0140000a (5/10) / buffer tsn: 数据文件对应的 tablespace 的 number 这只是dump文件中记录的数据而已 block 中是没有记录 tablespace 的 number 的 scn: 0x0000.0043890e seq: 0x05 flg: 0x02 tail: 0x890e0605frmt: 0x02 chkval: 0x0000 type: 0x06=trans dataBlock header dump: 0x0140000a Object id on Block? Y seg/obj: 0xd254 csc: 0x00.43890a itc: 2 flg: O typ: 1 - DATA fsl: 0 fnx: 0x0 ver: 0x01 Itl Xid Uba Flag Lck Scn/Fsc0x01 0x0004.00c.00001850 0x00801496.07b9.01 -U- 1 fsc 0x0000.0043890e0x02 0x0000.000.00000000 0x00000000.0000.00 - 0 fsc 0x0000.00000000 data_block_dump,data header at 0x87e125c / data_block_dump,data header at 0x87e125c 其实这个block不是直接从 data buffer 中 dump 出来的这个表示真正dump时 block 的数据区的起始位置 也就是下面这部分开始的位置 = / tsiz: hsiz: pbl: bdba: 在数据文件都是没有存储的 tsiz: 0x1fa0 / Total data area size 8k的block: 8192-20(block head)-24(Transaction Header)-24*2(一个事务条)-4(block tail)=8096(0x1fa0)hsiz: 0x14 / Data header size 数据块头20个字节+数据块尾4个字节=24字节(0x14)pbl: 0x087e125c / Pointer to buffer holding the blockbdba: 0x0140000a 76543210 flag=-ntab=1nrow=1frre=-1fsbo=0x14fseo=0x1f9bavsp=0x1f83tosp=0x1f830xe:pti0 nrow=1 offs=00x12:pri0 offs=0x1f9bblock_row_dump:tab 0, row 0, 0x1f9btl: 5 fb: -H-FL- lb: 0x1 cc: 1col 0: 1 61end_of_block_dumpEnd dump data blocks tsn: 17 file#: 5 minblk 10 maxblk 10block 坏掉了还可以报: ORA-600 (4519) Cache layer block type is incorrect ORA-600 (4393) Check for Type for Segment header with free list ORA-600 (4136) Check Rollback segment block ORA-600 (4154) Check Rollback segment block Ora-600kcbzpb_1,d,kind,chk gets signaled when the block got corrupted in memory. The only way it should be bad is if a stray store into memory destroyed the header or tail. d = blocknumber, kind= kind of corruption detected,chk = checksum flag ora-6003398 and ora-6003339 ora-6003398 is not in oracle 8. ora-6003398 means it failed a verification check before writing back to disk, so it must be an in-memory corruption. ora-6003339 comes with ora-1578 and means either disk corruption or in memory corruption after read. ora-600 3339 has been removed from 7.2+ From 7.2+ ora-600 3398 has become ora-600 3374 with some checks added.2进制存储格式 ALTER SESSION SET EVENTS 10289 trace name context forever, level 1; ALTER SESSION SET EVENTS 10289 trace name context off;为了回答一些疑问，引出本系列文章，讨论链接参考:/609499.html当我们使用file_hdrs事件来转储数据文件头信息时，Oracle会转储两部分信息，一部分来自控制文件，一部分来自数据文件，在数据库启动过程中，这两部分信息要用来进行启动验证。在数据库open的过程中,Oracle要进行检查中包含以下两个过程:第一次检查数据文件头中的C