Database Validation and Repair

By: Quinn Wildman

Abstract: This document describes exactly what gfix does when it attempts to validate or repair a database.

The following describes exactly what gfix does when it attempts to repair or validate a database. Note that this document is from the file VAL.C from the open source edition of InterBase 6.

Database Validation and Repair

  2. The following terminology will be helpful to understand in this discussion:

    record fragment: The smallest recognizable piece of a record; multiple fragments can be linked together to form a single version.
    record version: A single version of a record representing an INSERT, UPDATE or DELETE by a particular transaction (note that deletion of a record causes a new version to be stored as a
    deleted stub).
    record chain: A linked list of record versions chained together to represent a single logical "record".
    slot: The line number of the record on page. A variable-length array on each data page stores the offsets to the stored records on that page, and the slot is an index into that array.
  4. Here are all the options for gfix which have to do with validation, and what they do:

    gfix switch dpb parameter

    -validate isc_dpb_verify (gds__dpb_verify prior to 4.0)

    Invoke validation and repair. All other switches modify this switch.

    -full isc_dpb_records

    Visit all records. Without this switch, only page structures will be validated, which does involve some limited checking of records.

    -mend isc_dpb_repair

    Attempts to mend the database where it can to make it viable for reading; does not guarantee to retain data.

    -no_update isc_dpb_no_update

    Specifies that orphan pages not be released, and allocated pages not be marked in use when found to be free. Actually a misleading switch name since -mend will update the database, but if -mend is not specified and -no_update is specified, then no updates will occur to the database.

    -ignore isc_dpb_ignore

    Tells the engine to ignore checksums in fetching pages. Validate will report on the checksums, however. Should probably not even be a switch, it should just always be in effect. Otherwise checksums will disrupt the validation. Customers should be advised to always use it.

    NOTE:Unix 4.0 (ODS 8.0) does not have on-page checksums, and all platforms under ODS 9.0 (NevaStone & above) does not have checksums.
  6. Validation runs only with exclusive access to the database, to ensure that database structures are not modified during validation. On attach, validate attempts to obtain an exclusive lock on the database.

    If other attachments are already made locally or through the same multi- client server, validate gives up with the message:

    "Lock timeout during wait transaction
    -- Object "database_filename.gdb" is in use"

    If other processes or servers are attached to the database, validate waits for the exclusive lock on the database (i.e. waits for every other server to get out of the database).

    NOTE: Ordinarily when processes gain exclusive access to the database, all active transactions are marked as dead on the Transaction Inventory Pages. This feature is turned off for validation.

  8. There are two phases to the validation, the first of which is a walk through the entire database (described below). During this phase, all pages visited are stored in a bitmap for later use during the garbage collection phase.

    1. Visiting Pages
    2. During the walk-through phase, any page that is fetched goes through a basic validation:

      1. Page Type Check
      2. Each page is check against its expected type. If the wrong type page is found in the page header, the message:

        "Page xxx wrong type (expected xxx encountered xxx)"

        is returned. This could represent a) a problem with the database being overwritten, b) a bug with InterBase page allocation mechanisms in which one page was written over another, or c) a page which was allocated but never written to disk (most likely if the encountered page type was 0).

        The error does not tell you what page types are what, so here they are for reference:

        #define pag_undefined 0 /* purposely undefined */
        #define pag_header 1 /* Database header page */
        #define pag_pages 2 /* Page inventory page */
        #define pag_transactions 3 /* Transaction inventory page */
        #define pag_pointer 4 /* Pointer page */
        #define pag_data 5 /* Data page */
        #define pag_root 6 /* Index root page */
        #define pag_index 7 /* Index (B-tree) page */
        #define pag_blob 8 /* Blob data page */
        #define pag_ids 9 /* Gen-ids */
        #define pag_log 10 /* Write ahead log page: 4.0 only */

      3. Checksum
      4. If -ignore is specified, the checksum is specifically checked in validate instead of in the engine. If the checksum is found to be wrong, the error:

        "Checksum error on page xxx"

        is returned. This is harmless when found by validate, and the page will still continue to be validated--if data structures can be validated on page, they will be. If -mend is specified, the page will be marked for write, so that when the page is written to disk at the end of validation the checksum will automatically be recalculated.

        Note: For 4.0 only Windows & NLM platforms keep page checksums.

      5. Revisit
      6. We check each page fetched against the page bitmap to make sure we have not visited already. If we have, the error:

        "Page xxx doubly allocated"

        is returned. This should catch the case when a page of the same type is allocated for two different purposes.

        Data pages are not checked with the Revisit mechanism - when walking record chains and fragments they are frequently revisited.

    3. Garbage Collection
    4. During this phase, the Page Inventory (PIP) pages are checked against the bitmap of pages visited. Two types of errors can be detected during this phase.

      1. Orphan Pages
      2. If any pages in the page inventory were not visited during validation, the following error will be returned:

        "Page xxx is an orphan"

        If -no_update was not specified, the page will be marked as free on the PIP.

      3. Improperly Freed Pages
      4. If any pages marked free in the page inventory were in fact found to be in use during validation, the following error will be returned:

        "Page xxx is use but marked free" (sic)

        If -no_update was not specified, the page will be marked in use on the PIP.

        NOTE: If errors were found during the validation phase, no changes will be made to the PIP pages. This assumes that we did not have a chance to visit all the pages because invalid structures were detected.

    1. Page Fetching
    2. In order to ensure that all pages are fetched during validation, the following pages are fetched just for the most basic validation:

      1. The header page (and for 4.0 any overflow header pages).
      2. Log pages for after-image journalling (4.0 only).
      3. Page Inventory pages.
      4. Transaction Inventory pages
      5. If the system relation RDB$PAGES could not be read or did not contain any TIP pages, the message:

        "Transaction inventory pages lost"

        will be returned. If a particular page is missing from the sequence as established by RDB$PAGE_SEQUENCE, then the following message will be returned:

        "Transaction inventory page lost, sequence xxx"

        If -mend is specified, then a new TIP will be allocated on disk and stored in RDB$PAGES in the proper sequence. All transactions which would have been on that page are assumed committed.

        If a TIP page does not point to the next one in sequence, the following message will be returned:

        "Transaction inventory pages confused, sequence xxx"

      6. Generator pages as identified in RDB$PAGES.

    3. Relation Walking
    4. All the relations in the database are walked. For each relation, all indices defined on the relation are fetched, and all pointer and data pages associated with the relation are fetched (see below).

      But first, the metadata is scanned from RDB$RELATIONS to fetch the format of the relation. If this information is missing or corrupted the relation cannot be walked. If any bugchecks are encountered from the scan, the following message is returned:

      "bugcheck during scan of table xxx ()"

      This will prevent any further validation of the relation.

      NOTE: For views, the metadata is scanned but nothing further is done.

    5. Index Walking
    6. Prior to 4.5 (NevaStone) Indices were walked before data pages. In NevaStone Index walking was moved to after data page walking. Please refer to the later section entitled "Index Walking".

    7. Pointer Pages
    8. All the pointer pages for the relation are walked. As they are walked all child data pages are walked (see below). If a pointer page cannot be found, the following message is returned:

      "Pointer page (sequence xxx) lost"

      If the pointer page is not part of the relation we expected or if it is not marked as being in the proper sequence, the following message is returned:

      "Pointer page xxx is inconsistent"

      If each pointer page does not point to the next pointer page as stored in the RDB$PAGE_SEQUENCE field in RDB$PAGES, the following error is returned:

      "Pointer page (sequence xxx) inconsistent"

    9. Data Pages
    10. Each of the data pages referenced by the pointer page is fetched. If any are found to be corrupt at the page level, and -mend is specified, the page is deleted from its pointer page. This will cause a whole page of data to be lost.

      The data page is corrupt at the page level if it is not marked as part of the current relation, or if it is not marked as being in the proper sequence. If either of these conditions occurs, the following error is returned:

      "Data page xxx (sequence xxx) is confused"

    11. Slot Validation
    12. Each of the slots on the data page is looked at, up to the count of records stored on page. If the slot is non-zero, the record fragment at the specified offset is retrieved. If the record begins before the end of the slots array, or continues off the end of the page, the following error is returned:

      "Data page xxx (sequence xxx), line xxx is bad"

      where "line" means the slot number.

      NOTE: If this condition is encountered, the data page is considered corrupt at the page level (and thus will be removed from its pointer page if -mend is specified).

    13. Record Validation
    14. The record at each slot is looked at for basic validation, regardless of whether -full is specified or not. The fragment could be any of the following:

      1. Back Version
      2. If the fragment is marked as a back version, then it is skipped. It will be fetched as part of its record.

      3. Corrupt
      4. If the fragment is determined to be corrupt for any reason, and -mend is specified, then the record header is marked as damaged.

      5. Damaged
      6. If the fragment is marked damaged already from a previous visit or a previous validation, the following error is returned:

        "Record xxx is marked as damaged"

        where xxx is the record number.

      7. Bad Transaction
      8. If the record is marked with a transaction id greater than the last transaction started in the database, the following error is returned:

        "Record xxx has bad transaction xxx"

    15. Record Walking

      If -full is specified, and the fragment is the first fragment in a logical record, then the record at this slot number is fully retrieved. This involves retrieving all versions, and all fragments of each particular version. In other words, the entire logical record will be retrieved.

      1. Back Versions
      2. If there are any back versions, they are visited at this point. If the back version is on another page, the page is fetched but not validated since it will be walked separately.

        If the slot number of the back version is greater than the max records on page, or there is no record stored at that slot number, or it is a blob record, or it is a record fragment, or the fragment itself is invalid, the following error message is returned:

        "Chain for record xxx is broken"

      3. Incomplete
      4. If the record header is marked as incomplete, it means that there are additional fragments to be fetched--the record was too large to be stored in one slot. A pointer is stored in the record to the next fragment in the list.

        For fragmented records, all fragments are fetched to form a full record version. If any of the fragments is not in a valid position, or is not the correct length, the following error is returned:

        "Fragmented record xxx is corrupt"

        Once the full record has been retrieved, the length of the format is checked against the expected format stored in RDB$FORMATS (the format number is stored with the record, representing the exact format of the relation at the time the record was stored.) If the length of the reconstructed record does not match the expected format length, the following error is returned:

        "Record xxx is wrong length"

        For delta records (record versions which represent updates to the record) this check is not made.

    16. Blob Walking
    17. If the slot on the data page points to a blob record, then the blob is fetched (even without -full). This has several cases, corresponding to the various blob levels. (See the "Engine Internals" document for a discussion of blob levels.)

      Level Action
      0 These are just records on page, and no further validation is done.
      1 All the pages pointed to by the blob record are fetched and validated in sequence.
      2 All pages pointed to by the blob pointer pages are fetched and validated.
      3 The blob page is itself a blob pointer page; all its children are fetched and validated.

      For each blob page found, some further validation is done. If the page does not point back to the lead page, the following error is returned:

      "Warning: blob xxx appears inconsistent"

      where xxx corresponds to the blob record number. If any of the blob pages are not marked in the sequence we expect them to be in, the following error is returned:

      "Blob xxx is corrupt"

      Tip: the message for the same error in level 2 or 3 blobs is slightly different:

      "Blob xxx corrupt"

      If we have lost any of the blob pages in the sequence, the following error is returned:

      "Blob xxx is truncated"

      If the fetched blob is determined to be corrupt for any of the above reasons, and -mend is specified, then the blob record is marked as damaged.

    18. Index Walking
    19. In 4.5 (NevaStone) Index walking was moved to after the completion of data page walking.

      The indices for the relation are walked. If the index root page is missing, the following message is returned:

      "Missing index root page"

      and the indices are not walked. Otherwise the index root page is fetched and all indices on the page fetched.

      For each index, the btree pages are fetched from top-down, left to right. Basic validation is made on non-leaf pages to ensure that each node on page points to another index page. If -full validation is specified then the lower level page is fetched to ensure it is starting index entry is consistent with the parent entry. On leaf pages, the records pointed to by the index pages are not fetched, the keys are looked at to ensure they are in correct ascending order.

      If a visited page is not part of the specified relation and index, the following error is returned:

      "Index xxx is corrupt at page xxx"

      If there are orphan child pages, i.e. a child page does not have its entry as yet in the parent page, however the child's left sibling page has it's btr_sibling updated, the following error is returned

      "Index xxx has orphan child page at page xxx"

      If the page does not contain the number of nodes we would have expected from its marked length, the following error is returned:

      "Index xxx is corrupt on page xxx"

      While we are walking leaf pages, we keep a bitmap of all record numbers seen in the index. At the conclusion of the index walk we compare this bitmap to the bitmap of all records in the relation (calculated during data page/Record Validation phase). If the bitmaps are not equal then we have a corrupt index and the following error is reported:

      "Index %d is corrupt (missing entries)"

      We do NOT check that each version of each record has a valid index entry - nor do we check that the stored key for each item in the index corresponds to a version of the specified record.

    20. Relation Checking
    21. We count the number of backversions seen while walking pointer pages, and separately count the number of backversions seen while walking record chains. If these numbers do not match it indicates either "orphan" backversion chains or double-linked chains. If this is see the following error is returned:

      "Relation has %ld orphan backversions (%ld in use)"

      Currently we do not try to correct this condition, mearly report it. For "orphan" backversions the space can be reclaimed by a backup/restore. For double-linked chains a SWEEP should remove all the backversions.

    1. Damaged Records
    2. If any corruption of a record fragment is seen during validation, the record header is marked as "damaged". As far as I can see, this has no effect on the engine per se. Records marked as damaged will still be retrieved by the engine itself. There is some question in my mind as to whether this record should be retrieved at all during a gbak.

      If a damaged record is visited, the following error message will appear:

      "Record xxx is marked as damaged"

      Note that when a damaged record is first detected, this message is not actually printed. The record is simply marked as damaged. It is only thereafter when the record is visited that this message will appear. So I would postulate that unless a full validation is done at some point, you would not see this error message; once the full validation is done, the message will be returned even if you do not specify -full.

    3. Damaged Blobs
    4. Blob records marked as damaged cannot be opened and will not be deleted from disk. This means that even during backup the blob structures marked as damaged will not be fetched and backed up. (Why this is done differently for blobs than for records I cannot say. Perhaps it was viewed as too difficult to try to retrieve a damaged blob.)

Server Response from: ETNASC01