Baoding-FenShuaJiang
/
Green_GoodERP18_FSJ_Standard
關注 1
收藏 0
Wiki 捐赠
gooderp18绿色标准版
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
分支: master
分支列表 標籤
${ item.name }
建立分支 ${ searchTerm }
從 'master'
${ noResults }
Green_GoodERP18_FSJ_Standard/PostgreSQL/doc/postgresql/html/protocol-flow.html

894 line
60KB
原始文件 永久連結 Blame 歷史記錄 

  1. <?xml version="1.0" encoding="UTF-8" standalone="no"?>

  2. <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>52.2. Message Flow</title><link rel="stylesheet" type="text/css" href="stylesheet.css" /><link rev="made" href="pgsql-docs@lists.postgresql.org" /><meta name="generator" content="DocBook XSL Stylesheets V1.79.1" /><link rel="prev" href="protocol-overview.html" title="52.1. Overview" /><link rel="next" href="sasl-authentication.html" title="52.3. SASL Authentication" /></head><body><div xmlns="http://www.w3.org/TR/xhtml1/transitional" class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="5" align="center">52.2. Message Flow</th></tr><tr><td width="10%" align="left"><a accesskey="p" href="protocol-overview.html" title="52.1. Overview">Prev</a> </td><td width="10%" align="left"><a accesskey="u" href="protocol.html" title="Chapter 52. Frontend/Backend Protocol">Up</a></td><th width="60%" align="center">Chapter 52. Frontend/Backend Protocol</th><td width="10%" align="right"><a accesskey="h" href="index.html" title="PostgreSQL 12.4 Documentation">Home</a></td><td width="10%" align="right"> <a accesskey="n" href="sasl-authentication.html" title="52.3. SASL Authentication">Next</a></td></tr></table><hr></hr></div><div class="sect1" id="PROTOCOL-FLOW"><div class="titlepage"><div><div><h2 class="title" style="clear: both">52.2. Message Flow</h2></div></div></div><div class="toc"><dl class="toc"><dt><span class="sect2"><a href="protocol-flow.html#id-1.10.5.7.3">52.2.1. Start-up</a></span></dt><dt><span class="sect2"><a href="protocol-flow.html#id-1.10.5.7.4">52.2.2. Simple Query</a></span></dt><dt><span class="sect2"><a href="protocol-flow.html#PROTOCOL-FLOW-EXT-QUERY">52.2.3. Extended Query</a></span></dt><dt><span class="sect2"><a href="protocol-flow.html#id-1.10.5.7.6">52.2.4. Function Call</a></span></dt><dt><span class="sect2"><a href="protocol-flow.html#PROTOCOL-COPY">52.2.5. COPY Operations</a></span></dt><dt><span class="sect2"><a href="protocol-flow.html#PROTOCOL-ASYNC">52.2.6. Asynchronous Operations</a></span></dt><dt><span class="sect2"><a href="protocol-flow.html#id-1.10.5.7.9">52.2.7. Canceling Requests in Progress</a></span></dt><dt><span class="sect2"><a href="protocol-flow.html#id-1.10.5.7.10">52.2.8. Termination</a></span></dt><dt><span class="sect2"><a href="protocol-flow.html#id-1.10.5.7.11">52.2.9. <acronym class="acronym">SSL</acronym> Session Encryption</a></span></dt><dt><span class="sect2"><a href="protocol-flow.html#id-1.10.5.7.12">52.2.10. <acronym class="acronym">GSSAPI</acronym> Session Encryption</a></span></dt></dl></div><p>

  3.    This section describes the message flow and the semantics of each

  4.    message type.  (Details of the exact representation of each message

  5.    appear in <a class="xref" href="protocol-message-formats.html" title="52.7. Message Formats">Section 52.7</a>.)  There are

  6.    several different sub-protocols depending on the state of the

  7.    connection: start-up, query, function call,

  8.    <code class="command">COPY</code>, and termination.  There are also special

  9.    provisions for asynchronous operations (including notification

  10.    responses and command cancellation), which can occur at any time

  11.    after the start-up phase.

  12.   </p><div class="sect2" id="id-1.10.5.7.3"><div class="titlepage"><div><div><h3 class="title">52.2.1. Start-up</h3></div></div></div><p>

  13.     To begin a session, a frontend opens a connection to the server and sends

  14.     a startup message.  This message includes the names of the user and of the

  15.     database the user wants to connect to; it also identifies the particular

  16.     protocol version to be used.  (Optionally, the startup message can include

  17.     additional settings for run-time parameters.)

  18.     The server then uses this information and

  19.     the contents of its configuration files (such as

  20.     <code class="filename">pg_hba.conf</code>) to determine

  21.     whether the connection is provisionally acceptable, and what additional

  22.     authentication is required (if any).

  23.    </p><p>

  24.     The server then sends an appropriate authentication request message,

  25.     to which the frontend must reply with an appropriate authentication

  26.     response message (such as a password).

  27.     For all authentication methods except GSSAPI, SSPI and SASL, there is at

  28.     most one request and one response. In some methods, no response

  29.     at all is needed from the frontend, and so no authentication request

  30.     occurs. For GSSAPI, SSPI and SASL, multiple exchanges of packets may be

  31.     needed to complete the authentication.

  32.    </p><p>

  33.     The authentication cycle ends with the server either rejecting the

  34.     connection attempt (ErrorResponse), or sending AuthenticationOk.

  35.    </p><p>

  36.     The possible messages from the server in this phase are:

  37. 

  38.     </p><div class="variablelist"><dl class="variablelist"><dt><span class="term">ErrorResponse</span></dt><dd><p>

  39.         The connection attempt has been rejected.

  40.         The server then immediately closes the connection.

  41.        </p></dd><dt><span class="term">AuthenticationOk</span></dt><dd><p>

  42.         The authentication exchange is successfully completed.

  43.        </p></dd><dt><span class="term">AuthenticationKerberosV5</span></dt><dd><p>

  44.         The frontend must now take part in a Kerberos V5

  45.         authentication dialog (not described here, part of the

  46.         Kerberos specification) with the server.  If this is

  47.         successful, the server responds with an AuthenticationOk,

  48.         otherwise it responds with an ErrorResponse. This is no

  49.         longer supported.

  50.        </p></dd><dt><span class="term">AuthenticationCleartextPassword</span></dt><dd><p>

  51.         The frontend must now send a PasswordMessage containing the

  52.         password in clear-text form.  If

  53.         this is the correct password, the server responds with an

  54.         AuthenticationOk, otherwise it responds with an ErrorResponse.

  55.        </p></dd><dt><span class="term">AuthenticationMD5Password</span></dt><dd><p>

  56.         The frontend must now send a PasswordMessage containing the

  57.         password (with user name) encrypted via MD5, then encrypted

  58.         again using the 4-byte random salt specified in the

  59.         AuthenticationMD5Password message.  If this is the correct

  60.         password, the server responds with an AuthenticationOk,

  61.         otherwise it responds with an ErrorResponse.  The actual

  62.         PasswordMessage can be computed in SQL as <code class="literal">concat('md5',

  63.         md5(concat(md5(concat(password, username)), random-salt)))</code>.

  64.         (Keep in mind the <code class="function">md5()</code> function returns its

  65.         result as a hex string.)

  66.        </p></dd><dt><span class="term">AuthenticationSCMCredential</span></dt><dd><p>

  67.         This response is only possible for local Unix-domain connections

  68.         on platforms that support SCM credential messages.  The frontend

  69.         must issue an SCM credential message and then send a single data

  70.         byte.  (The contents of the data byte are uninteresting; it's

  71.         only used to ensure that the server waits long enough to receive

  72.         the credential message.)  If the credential is acceptable,

  73.         the server responds with an

  74.         AuthenticationOk, otherwise it responds with an ErrorResponse.

  75.         (This message type is only issued by pre-9.1 servers.  It may

  76.         eventually be removed from the protocol specification.)

  77.        </p></dd><dt><span class="term">AuthenticationGSS</span></dt><dd><p>

  78.         The frontend must now initiate a GSSAPI negotiation. The frontend

  79.         will send a GSSResponse message with the first part of the GSSAPI

  80.         data stream in response to this. If further messages are needed,

  81.         the server will respond with AuthenticationGSSContinue.

  82.        </p></dd><dt><span class="term">AuthenticationSSPI</span></dt><dd><p>

  83.         The frontend must now initiate a SSPI negotiation. The frontend

  84.         will send a GSSResponse with the first part of the SSPI

  85.         data stream in response to this. If further messages are needed,

  86.         the server will respond with AuthenticationGSSContinue.

  87.        </p></dd><dt><span class="term">AuthenticationGSSContinue</span></dt><dd><p>

  88.         This message contains the response data from the previous step

  89.         of GSSAPI or SSPI negotiation (AuthenticationGSS, AuthenticationSSPI

  90.         or a previous AuthenticationGSSContinue). If the GSSAPI

  91.         or SSPI data in this message

  92.         indicates more data is needed to complete the authentication,

  93.         the frontend must send that data as another GSSResponse message. If

  94.         GSSAPI or SSPI authentication is completed by this message, the server

  95.         will next send AuthenticationOk to indicate successful authentication

  96.         or ErrorResponse to indicate failure.

  97.        </p></dd><dt><span class="term">AuthenticationSASL</span></dt><dd><p>

  98.         The frontend must now initiate a SASL negotiation, using one of the

  99.         SASL mechanisms listed in the message. The frontend will send a

  100.         SASLInitialResponse with the name of the selected mechanism, and the

  101.         first part of the SASL data stream in response to this. If further

  102.         messages are needed, the server will respond with

  103.         AuthenticationSASLContinue. See <a class="xref" href="sasl-authentication.html" title="52.3. SASL Authentication">Section 52.3</a>

  104.         for details.

  105.        </p></dd><dt><span class="term">AuthenticationSASLContinue</span></dt><dd><p>

  106.         This message contains challenge data from the previous step of SASL

  107.         negotiation (AuthenticationSASL, or a previous

  108.         AuthenticationSASLContinue). The frontend must respond with a

  109.         SASLResponse message.

  110.        </p></dd><dt><span class="term">AuthenticationSASLFinal</span></dt><dd><p>

  111.         SASL authentication has completed with additional mechanism-specific

  112.         data for the client. The server will next send AuthenticationOk to

  113.         indicate successful authentication, or an ErrorResponse to indicate

  114.         failure. This message is sent only if the SASL mechanism specifies

  115.         additional data to be sent from server to client at completion.

  116.        </p></dd><dt><span class="term">NegotiateProtocolVersion</span></dt><dd><p>

  117.         The server does not support the minor protocol version requested

  118.         by the client, but does support an earlier version of the protocol;

  119.         this message indicates the highest supported minor version.  This

  120.         message will also be sent if the client requested unsupported protocol

  121.         options (i.e. beginning with <code class="literal">_pq_.</code>) in the

  122.         startup packet.  This message will be followed by an ErrorResponse or

  123.         a message indicating the success or failure of authentication.

  124.        </p></dd></dl></div><p>

  125.    </p><p>

  126.     If the frontend does not support the authentication method

  127.     requested by the server, then it should immediately close the

  128.     connection.

  129.    </p><p>

  130.     After having received AuthenticationOk, the frontend must wait

  131.     for further messages from the server.  In this phase a backend process

  132.     is being started, and the frontend is just an interested bystander.

  133.     It is still possible for the startup attempt

  134.     to fail (ErrorResponse) or the server to decline support for the requested

  135.     minor protocol version (NegotiateProtocolVersion), but in the normal case

  136.     the backend will send some ParameterStatus messages, BackendKeyData, and

  137.     finally ReadyForQuery.

  138.    </p><p>

  139.     During this phase the backend will attempt to apply any additional

  140.     run-time parameter settings that were given in the startup message.

  141.     If successful, these values become session defaults.  An error causes

  142.     ErrorResponse and exit.

  143.    </p><p>

  144.     The possible messages from the backend in this phase are:

  145. 

  146.     </p><div class="variablelist"><dl class="variablelist"><dt><span class="term">BackendKeyData</span></dt><dd><p>

  147.         This message provides secret-key data that the frontend must

  148.         save if it wants to be able to issue cancel requests later.

  149.         The frontend should not respond to this message, but should

  150.         continue listening for a ReadyForQuery message.

  151.        </p></dd><dt><span class="term">ParameterStatus</span></dt><dd><p>

  152.         This message informs the frontend about the current (initial)

  153.          setting of backend parameters, such as <a class="xref" href="runtime-config-client.html#GUC-CLIENT-ENCODING">client_encoding</a> or <a class="xref" href="runtime-config-client.html#GUC-DATESTYLE">DateStyle</a>.

  154.          The frontend can ignore this message, or record the settings

  155.          for its future use; see <a class="xref" href="protocol-flow.html#PROTOCOL-ASYNC" title="52.2.6. Asynchronous Operations">Section 52.2.6</a> for

  156.          more details.  The frontend should not respond to this

  157.          message, but should continue listening for a ReadyForQuery

  158.          message.

  159.        </p></dd><dt><span class="term">ReadyForQuery</span></dt><dd><p>

  160.         Start-up is completed.  The frontend can now issue commands.

  161.        </p></dd><dt><span class="term">ErrorResponse</span></dt><dd><p>

  162.         Start-up failed.  The connection is closed after sending this

  163.         message.

  164.        </p></dd><dt><span class="term">NoticeResponse</span></dt><dd><p>

  165.         A warning message has been issued.  The frontend should

  166.         display the message but continue listening for ReadyForQuery

  167.         or ErrorResponse.

  168.        </p></dd></dl></div><p>

  169.    </p><p>

  170.     The ReadyForQuery message is the same one that the backend will

  171.     issue after each command cycle.  Depending on the coding needs of

  172.     the frontend, it is reasonable to consider ReadyForQuery as

  173.     starting a command cycle, or to consider ReadyForQuery as ending the

  174.     start-up phase and each subsequent command cycle.

  175.    </p></div><div class="sect2" id="id-1.10.5.7.4"><div class="titlepage"><div><div><h3 class="title">52.2.2. Simple Query</h3></div></div></div><p>

  176.     A simple query cycle is initiated by the frontend sending a Query message

  177.     to the backend.  The message includes an SQL command (or commands)

  178.     expressed as a text string.

  179.     The backend then sends one or more response

  180.     messages depending on the contents of the query command string,

  181.     and finally a ReadyForQuery response message.  ReadyForQuery

  182.     informs the frontend that it can safely send a new command.

  183.     (It is not actually necessary for the frontend to wait for

  184.     ReadyForQuery before issuing another command, but the frontend must

  185.     then take responsibility for figuring out what happens if the earlier

  186.     command fails and already-issued later commands succeed.)

  187.    </p><p>

  188.     The possible response messages from the backend are:

  189. 

  190.     </p><div class="variablelist"><dl class="variablelist"><dt><span class="term">CommandComplete</span></dt><dd><p>

  191.         An SQL command completed normally.

  192.        </p></dd><dt><span class="term">CopyInResponse</span></dt><dd><p>

  193.         The backend is ready to copy data from the frontend to a

  194.         table; see <a class="xref" href="protocol-flow.html#PROTOCOL-COPY" title="52.2.5. COPY Operations">Section 52.2.5</a>.

  195.        </p></dd><dt><span class="term">CopyOutResponse</span></dt><dd><p>

  196.         The backend is ready to copy data from a table to the

  197.         frontend; see <a class="xref" href="protocol-flow.html#PROTOCOL-COPY" title="52.2.5. COPY Operations">Section 52.2.5</a>.

  198.        </p></dd><dt><span class="term">RowDescription</span></dt><dd><p>

  199.         Indicates that rows are about to be returned in response to

  200.         a <code class="command">SELECT</code>, <code class="command">FETCH</code>, etc query.

  201.         The contents of this message describe the column layout of the rows.

  202.         This will be followed by a DataRow message for each row being returned

  203.         to the frontend.

  204.        </p></dd><dt><span class="term">DataRow</span></dt><dd><p>

  205.         One of the set of rows returned by

  206.         a <code class="command">SELECT</code>, <code class="command">FETCH</code>, etc query.

  207.        </p></dd><dt><span class="term">EmptyQueryResponse</span></dt><dd><p>

  208.         An empty query string was recognized.

  209.        </p></dd><dt><span class="term">ErrorResponse</span></dt><dd><p>

  210.         An error has occurred.

  211.        </p></dd><dt><span class="term">ReadyForQuery</span></dt><dd><p>

  212.         Processing of the query string is complete.  A separate

  213.         message is sent to indicate this because the query string might

  214.         contain multiple SQL commands.  (CommandComplete marks the

  215.         end of processing one SQL command, not the whole string.)

  216.         ReadyForQuery will always be sent, whether processing

  217.         terminates successfully or with an error.

  218.        </p></dd><dt><span class="term">NoticeResponse</span></dt><dd><p>

  219.         A warning message has been issued in relation to the query.

  220.         Notices are in addition to other responses, i.e., the backend

  221.         will continue processing the command.

  222.        </p></dd></dl></div><p>

  223.    </p><p>

  224.     The response to a <code class="command">SELECT</code> query (or other queries that

  225.     return row sets, such as <code class="command">EXPLAIN</code> or <code class="command">SHOW</code>)

  226.     normally consists of RowDescription, zero or more

  227.     DataRow messages, and then CommandComplete.

  228.     <code class="command">COPY</code> to or from the frontend invokes special protocol

  229.     as described in <a class="xref" href="protocol-flow.html#PROTOCOL-COPY" title="52.2.5. COPY Operations">Section 52.2.5</a>.

  230.     All other query types normally produce only

  231.     a CommandComplete message.

  232.    </p><p>

  233.     Since a query string could contain several queries (separated by

  234.     semicolons), there might be several such response sequences before the

  235.     backend finishes processing the query string.  ReadyForQuery is issued

  236.     when the entire string has been processed and the backend is ready to

  237.     accept a new query string.

  238.    </p><p>

  239.     If a completely empty (no contents other than whitespace) query string

  240.     is received, the response is EmptyQueryResponse followed by ReadyForQuery.

  241.    </p><p>

  242.     In the event of an error, ErrorResponse is issued followed by

  243.     ReadyForQuery.  All further processing of the query string is aborted by

  244.     ErrorResponse (even if more queries remained in it).  Note that this

  245.     might occur partway through the sequence of messages generated by an

  246.     individual query.

  247.    </p><p>

  248.     In simple Query mode, the format of retrieved values is always text,

  249.     except when the given command is a <code class="command">FETCH</code> from a cursor

  250.     declared with the <code class="literal">BINARY</code> option.  In that case, the

  251.     retrieved values are in binary format.  The format codes given in

  252.     the RowDescription message tell which format is being used.

  253.    </p><p>

  254.     A frontend must be prepared to accept ErrorResponse and

  255.     NoticeResponse messages whenever it is expecting any other type of

  256.     message.  See also <a class="xref" href="protocol-flow.html#PROTOCOL-ASYNC" title="52.2.6. Asynchronous Operations">Section 52.2.6</a> concerning messages

  257.     that the backend might generate due to outside events.

  258.    </p><p>

  259.     Recommended practice is to code frontends in a state-machine style

  260.     that will accept any message type at any time that it could make sense,

  261.     rather than wiring in assumptions about the exact sequence of messages.

  262.    </p><div class="sect3" id="PROTOCOL-FLOW-MULTI-STATEMENT"><div class="titlepage"><div><div><h4 class="title">52.2.2.1. Multiple Statements in a Simple Query</h4></div></div></div><p>

  263.      When a simple Query message contains more than one SQL statement

  264.      (separated by semicolons), those statements are executed as a single

  265.      transaction, unless explicit transaction control commands are included

  266.      to force a different behavior.  For example, if the message contains

  267. </p><pre class="programlisting">

  268. INSERT INTO mytable VALUES(1);

  269. SELECT 1/0;

  270. INSERT INTO mytable VALUES(2);

  271. </pre><p>

  272.      then the divide-by-zero failure in the <code class="command">SELECT</code> will force

  273.      rollback of the first <code class="command">INSERT</code>.  Furthermore, because

  274.      execution of the message is abandoned at the first error, the second

  275.      <code class="command">INSERT</code> is never attempted at all.

  276.     </p><p>

  277.      If instead the message contains

  278. </p><pre class="programlisting">

  279. BEGIN;

  280. INSERT INTO mytable VALUES(1);

  281. COMMIT;

  282. INSERT INTO mytable VALUES(2);

  283. SELECT 1/0;

  284. </pre><p>

  285.      then the first <code class="command">INSERT</code> is committed by the

  286.      explicit <code class="command">COMMIT</code> command.  The second <code class="command">INSERT</code>

  287.      and the <code class="command">SELECT</code> are still treated as a single transaction,

  288.      so that the divide-by-zero failure will roll back the

  289.      second <code class="command">INSERT</code>, but not the first one.

  290.     </p><p>

  291.      This behavior is implemented by running the statements in a

  292.      multi-statement Query message in an <em class="firstterm">implicit transaction

  293.      block</em> unless there is some explicit transaction block for them to

  294.      run in.  The main difference between an implicit transaction block and

  295.      a regular one is that an implicit block is closed automatically at the

  296.      end of the Query message, either by an implicit commit if there was no

  297.      error, or an implicit rollback if there was an error.  This is similar

  298.      to the implicit commit or rollback that happens for a statement

  299.      executed by itself (when not in a transaction block).

  300.     </p><p>

  301.      If the session is already in a transaction block, as a result of

  302.      a <code class="command">BEGIN</code> in some previous message, then the Query message

  303.      simply continues that transaction block, whether the message contains

  304.      one statement or several.  However, if the Query message contains

  305.      a <code class="command">COMMIT</code> or <code class="command">ROLLBACK</code> closing the existing

  306.      transaction block, then any following statements are executed in an

  307.      implicit transaction block.

  308.      Conversely, if a <code class="command">BEGIN</code> appears in a multi-statement Query

  309.      message, then it starts a regular transaction block that will only be

  310.      terminated by an explicit <code class="command">COMMIT</code> or <code class="command">ROLLBACK</code>,

  311.      whether that appears in this Query message or a later one.

  312.      If the <code class="command">BEGIN</code> follows some statements that were executed as

  313.      an implicit transaction block, those statements are not immediately

  314.      committed; in effect, they are retroactively included into the new

  315.      regular transaction block.

  316.     </p><p>

  317.      A <code class="command">COMMIT</code> or <code class="command">ROLLBACK</code> appearing in an implicit

  318.      transaction block is executed as normal, closing the implicit block;

  319.      however, a warning will be issued since a <code class="command">COMMIT</code>

  320.      or <code class="command">ROLLBACK</code> without a previous <code class="command">BEGIN</code> might

  321.      represent a mistake.  If more statements follow, a new implicit

  322.      transaction block will be started for them.

  323.     </p><p>

  324.      Savepoints are not allowed in an implicit transaction block, since

  325.      they would conflict with the behavior of automatically closing the

  326.      block upon any error.

  327.     </p><p>

  328.      Remember that, regardless of any transaction control commands that may

  329.      be present, execution of the Query message stops at the first error.

  330.      Thus for example given

  331. </p><pre class="programlisting">

  332. BEGIN;

  333. SELECT 1/0;

  334. ROLLBACK;

  335. </pre><p>

  336.      in a single Query message, the session will be left inside a failed

  337.      regular transaction block, since the <code class="command">ROLLBACK</code> is not

  338.      reached after the divide-by-zero error.  Another <code class="command">ROLLBACK</code>

  339.      will be needed to restore the session to a usable state.

  340.     </p><p>

  341.      Another behavior of note is that initial lexical and syntactic

  342.      analysis is done on the entire query string before any of it is

  343.      executed.  Thus simple errors (such as a misspelled keyword) in later

  344.      statements can prevent execution of any of the statements.  This

  345.      is normally invisible to users since the statements would all roll

  346.      back anyway when done as an implicit transaction block.  However,

  347.      it can be visible when attempting to do multiple transactions within a

  348.      multi-statement Query.  For instance, if a typo turned our previous

  349.      example into

  350. </p><pre class="programlisting">

  351. BEGIN;

  352. INSERT INTO mytable VALUES(1);

  353. COMMIT;

  354. INSERT INTO mytable VALUES(2);

  355. SELCT 1/0;

  356. </pre><p>

  357.      then none of the statements would get run, resulting in the visible

  358.      difference that the first <code class="command">INSERT</code> is not committed.

  359.      Errors detected at semantic analysis or later, such as a misspelled

  360.      table or column name, do not have this effect.

  361.     </p></div></div><div class="sect2" id="PROTOCOL-FLOW-EXT-QUERY"><div class="titlepage"><div><div><h3 class="title">52.2.3. Extended Query</h3></div></div></div><p>

  362.     The extended query protocol breaks down the above-described simple

  363.     query protocol into multiple steps.  The results of preparatory

  364.     steps can be re-used multiple times for improved efficiency.

  365.     Furthermore, additional features are available, such as the possibility

  366.     of supplying data values as separate parameters instead of having to

  367.     insert them directly into a query string.

  368.    </p><p>

  369.     In the extended protocol, the frontend first sends a Parse message,

  370.     which contains a textual query string, optionally some information

  371.     about data types of parameter placeholders, and the

  372.     name of a destination prepared-statement object (an empty string

  373.     selects the unnamed prepared statement).  The response is

  374.     either ParseComplete or ErrorResponse.  Parameter data types can be

  375.     specified by OID; if not given, the parser attempts to infer the

  376.     data types in the same way as it would do for untyped literal string

  377.     constants.

  378.    </p><div class="note"><h3 class="title">Note</h3><p>

  379.      A parameter data type can be left unspecified by setting it to zero,

  380.      or by making the array of parameter type OIDs shorter than the

  381.      number of parameter symbols (<code class="literal">$</code><em class="replaceable"><code>n</code></em>)

  382.      used in the query string.  Another special case is that a parameter's

  383.      type can be specified as <code class="type">void</code> (that is, the OID of the

  384.      <code class="type">void</code> pseudo-type).  This is meant to allow parameter symbols

  385.      to be used for function parameters that are actually OUT parameters.

  386.      Ordinarily there is no context in which a <code class="type">void</code> parameter

  387.      could be used, but if such a parameter symbol appears in a function's

  388.      parameter list, it is effectively ignored.  For example, a function

  389.      call such as <code class="literal">foo($1,$2,$3,$4)</code> could match a function with

  390.      two IN and two OUT arguments, if <code class="literal">$3</code> and <code class="literal">$4</code>

  391.      are specified as having type <code class="type">void</code>.

  392.     </p></div><div class="note"><h3 class="title">Note</h3><p>

  393.      The query string contained in a Parse message cannot include more

  394.      than one SQL statement; else a syntax error is reported.  This

  395.      restriction does not exist in the simple-query protocol, but it

  396.      does exist in the extended protocol, because allowing prepared

  397.      statements or portals to contain multiple commands would complicate

  398.      the protocol unduly.

  399.     </p></div><p>

  400.     If successfully created, a named prepared-statement object lasts till

  401.     the end of the current session, unless explicitly destroyed.  An unnamed

  402.     prepared statement lasts only until the next Parse statement specifying

  403.     the unnamed statement as destination is issued.  (Note that a simple

  404.     Query message also destroys the unnamed statement.)  Named prepared

  405.     statements must be explicitly closed before they can be redefined by

  406.     another Parse message, but this is not required for the unnamed statement.

  407.     Named prepared statements can also be created and accessed at the SQL

  408.     command level, using <code class="command">PREPARE</code> and <code class="command">EXECUTE</code>.

  409.    </p><p>

  410.     Once a prepared statement exists, it can be readied for execution using a

  411.     Bind message.  The Bind message gives the name of the source prepared

  412.     statement (empty string denotes the unnamed prepared statement), the name

  413.     of the destination portal (empty string denotes the unnamed portal), and

  414.     the values to use for any parameter placeholders present in the prepared

  415.     statement.  The

  416.     supplied parameter set must match those needed by the prepared statement.

  417.     (If you declared any <code class="type">void</code> parameters in the Parse message,

  418.     pass NULL values for them in the Bind message.)

  419.     Bind also specifies the format to use for any data returned

  420.     by the query; the format can be specified overall, or per-column.

  421.     The response is either BindComplete or ErrorResponse.

  422.    </p><div class="note"><h3 class="title">Note</h3><p>

  423.      The choice between text and binary output is determined by the format

  424.      codes given in Bind, regardless of the SQL command involved.  The

  425.      <code class="literal">BINARY</code> attribute in cursor declarations is irrelevant when

  426.      using extended query protocol.

  427.     </p></div><p>

  428.     Query planning typically occurs when the Bind message is processed.

  429.     If the prepared statement has no parameters, or is executed repeatedly,

  430.     the server might save the created plan and re-use it during subsequent

  431.     Bind messages for the same prepared statement.  However, it will do so

  432.     only if it finds that a generic plan can be created that is not much

  433.     less efficient than a plan that depends on the specific parameter values

  434.     supplied.  This happens transparently so far as the protocol is concerned.

  435.    </p><p>

  436.     If successfully created, a named portal object lasts till the end of the

  437.     current transaction, unless explicitly destroyed.  An unnamed portal is

  438.     destroyed at the end of the transaction, or as soon as the next Bind

  439.     statement specifying the unnamed portal as destination is issued.  (Note

  440.     that a simple Query message also destroys the unnamed portal.)  Named

  441.     portals must be explicitly closed before they can be redefined by another

  442.     Bind message, but this is not required for the unnamed portal.

  443.     Named portals can also be created and accessed at the SQL

  444.     command level, using <code class="command">DECLARE CURSOR</code> and <code class="command">FETCH</code>.

  445.    </p><p>

  446.     Once a portal exists, it can be executed using an Execute message.

  447.     The Execute message specifies the portal name (empty string denotes the

  448.     unnamed portal) and

  449.     a maximum result-row count (zero meaning <span class="quote">“<span class="quote">fetch all rows</span>”</span>).

  450.     The result-row count is only meaningful for portals

  451.     containing commands that return row sets; in other cases the command is

  452.     always executed to completion, and the row count is ignored.

  453.     The possible

  454.     responses to Execute are the same as those described above for queries

  455.     issued via simple query protocol, except that Execute doesn't cause

  456.     ReadyForQuery or RowDescription to be issued.

  457.    </p><p>

  458.     If Execute terminates before completing the execution of a portal

  459.     (due to reaching a nonzero result-row count), it will send a

  460.     PortalSuspended message; the appearance of this message tells the frontend

  461.     that another Execute should be issued against the same portal to

  462.     complete the operation.  The CommandComplete message indicating

  463.     completion of the source SQL command is not sent until

  464.     the portal's execution is completed.  Therefore, an Execute phase is

  465.     always terminated by the appearance of exactly one of these messages:

  466.     CommandComplete, EmptyQueryResponse (if the portal was created from

  467.     an empty query string), ErrorResponse, or PortalSuspended.

  468.    </p><p>

  469.     At completion of each series of extended-query messages, the frontend

  470.     should issue a Sync message.  This parameterless message causes the

  471.     backend to close the current transaction if it's not inside a

  472.     <code class="command">BEGIN</code>/<code class="command">COMMIT</code> transaction block (<span class="quote">“<span class="quote">close</span>”</span>

  473.     meaning to commit if no error, or roll back if error).  Then a

  474.     ReadyForQuery response is issued.  The purpose of Sync is to provide

  475.     a resynchronization point for error recovery.  When an error is detected

  476.     while processing any extended-query message, the backend issues

  477.     ErrorResponse, then reads and discards messages until a Sync is reached,

  478.     then issues ReadyForQuery and returns to normal message processing.

  479.     (But note that no skipping occurs if an error is detected

  480.     <span class="emphasis"><em>while</em></span> processing Sync — this ensures that there is one

  481.     and only one ReadyForQuery sent for each Sync.)

  482.    </p><div class="note"><h3 class="title">Note</h3><p>

  483.      Sync does not cause a transaction block opened with <code class="command">BEGIN</code>

  484.      to be closed.  It is possible to detect this situation since the

  485.      ReadyForQuery message includes transaction status information.

  486.     </p></div><p>

  487.     In addition to these fundamental, required operations, there are several

  488.     optional operations that can be used with extended-query protocol.

  489.    </p><p>

  490.     The Describe message (portal variant) specifies the name of an existing

  491.     portal (or an empty string for the unnamed portal).  The response is a

  492.     RowDescription message describing the rows that will be returned by

  493.     executing the portal; or a NoData message if the portal does not contain a

  494.     query that will return rows; or ErrorResponse if there is no such portal.

  495.    </p><p>

  496.     The Describe message (statement variant) specifies the name of an existing

  497.     prepared statement (or an empty string for the unnamed prepared

  498.     statement).  The response is a ParameterDescription message describing the

  499.     parameters needed by the statement, followed by a RowDescription message

  500.     describing the rows that will be returned when the statement is eventually

  501.     executed (or a NoData message if the statement will not return rows).

  502.     ErrorResponse is issued if there is no such prepared statement.  Note that

  503.     since Bind has not yet been issued, the formats to be used for returned

  504.     columns are not yet known to the backend; the format code fields in the

  505.     RowDescription message will be zeroes in this case.

  506.    </p><div class="tip"><h3 class="title">Tip</h3><p>

  507.      In most scenarios the frontend should issue one or the other variant

  508.      of Describe before issuing Execute, to ensure that it knows how to

  509.      interpret the results it will get back.

  510.     </p></div><p>

  511.     The Close message closes an existing prepared statement or portal

  512.     and releases resources.  It is not an error to issue Close against

  513.     a nonexistent statement or portal name.  The response is normally

  514.     CloseComplete, but could be ErrorResponse if some difficulty is

  515.     encountered while releasing resources.  Note that closing a prepared

  516.     statement implicitly closes any open portals that were constructed

  517.     from that statement.

  518.    </p><p>

  519.     The Flush message does not cause any specific output to be generated,

  520.     but forces the backend to deliver any data pending in its output

  521.     buffers.  A Flush must be sent after any extended-query command except

  522.     Sync, if the frontend wishes to examine the results of that command before

  523.     issuing more commands.  Without Flush, messages returned by the backend

  524.     will be combined into the minimum possible number of packets to minimize

  525.     network overhead.

  526.    </p><div class="note"><h3 class="title">Note</h3><p>

  527.      The simple Query message is approximately equivalent to the series Parse,

  528.      Bind, portal Describe, Execute, Close, Sync, using the unnamed prepared

  529.      statement and portal objects and no parameters.  One difference is that

  530.      it will accept multiple SQL statements in the query string, automatically

  531.      performing the bind/describe/execute sequence for each one in succession.

  532.      Another difference is that it will not return ParseComplete, BindComplete,

  533.      CloseComplete, or NoData messages.

  534.     </p></div></div><div class="sect2" id="id-1.10.5.7.6"><div class="titlepage"><div><div><h3 class="title">52.2.4. Function Call</h3></div></div></div><p>

  535.     The Function Call sub-protocol allows the client to request a direct

  536.     call of any function that exists in the database's

  537.     <code class="structname">pg_proc</code> system catalog.  The client must have

  538.     execute permission for the function.

  539.    </p><div class="note"><h3 class="title">Note</h3><p>

  540.      The Function Call sub-protocol is a legacy feature that is probably best

  541.      avoided in new code.  Similar results can be accomplished by setting up

  542.      a prepared statement that does <code class="literal">SELECT function($1, ...)</code>.

  543.      The Function Call cycle can then be replaced with Bind/Execute.

  544.     </p></div><p>

  545.     A Function Call cycle is initiated by the frontend sending a

  546.     FunctionCall message to the backend.  The backend then sends one

  547.     or more response messages depending on the results of the function

  548.     call, and finally a ReadyForQuery response message.  ReadyForQuery

  549.     informs the frontend that it can safely send a new query or

  550.     function call.

  551.    </p><p>

  552.     The possible response messages from the backend are:

  553. 

  554.     </p><div class="variablelist"><dl class="variablelist"><dt><span class="term">ErrorResponse</span></dt><dd><p>

  555.         An error has occurred.

  556.        </p></dd><dt><span class="term">FunctionCallResponse</span></dt><dd><p>

  557.         The function call was completed and returned the result given

  558.         in the message.

  559.         (Note that the Function Call protocol can only handle a single

  560.         scalar result, not a row type or set of results.)

  561.        </p></dd><dt><span class="term">ReadyForQuery</span></dt><dd><p>

  562.         Processing of the function call is complete.  ReadyForQuery

  563.         will always be sent, whether processing terminates

  564.         successfully or with an error.

  565.        </p></dd><dt><span class="term">NoticeResponse</span></dt><dd><p>

  566.         A warning message has been issued in relation to the function

  567.         call.  Notices are in addition to other responses, i.e., the

  568.         backend will continue processing the command.

  569.        </p></dd></dl></div><p>

  570.    </p></div><div class="sect2" id="PROTOCOL-COPY"><div class="titlepage"><div><div><h3 class="title">52.2.5. COPY Operations</h3></div></div></div><p>

  571.     The <code class="command">COPY</code> command allows high-speed bulk data transfer

  572.     to or from the server.  Copy-in and copy-out operations each switch

  573.     the connection into a distinct sub-protocol, which lasts until the

  574.     operation is completed.

  575.    </p><p>

  576.     Copy-in mode (data transfer to the server) is initiated when the

  577.     backend executes a <code class="command">COPY FROM STDIN</code> SQL statement.  The backend

  578.     sends a CopyInResponse message to the frontend.  The frontend should

  579.     then send zero or more CopyData messages, forming a stream of input

  580.     data.  (The message boundaries are not required to have anything to do

  581.     with row boundaries, although that is often a reasonable choice.)

  582.     The frontend can terminate the copy-in mode by sending either a CopyDone

  583.     message (allowing successful termination) or a CopyFail message (which

  584.     will cause the <code class="command">COPY</code> SQL statement to fail with an

  585.     error).  The backend then reverts to the command-processing mode it was

  586.     in before the <code class="command">COPY</code> started, which will be either simple or

  587.     extended query protocol.  It will next send either CommandComplete

  588.     (if successful) or ErrorResponse (if not).

  589.    </p><p>

  590.     In the event of a backend-detected error during copy-in mode (including

  591.     receipt of a CopyFail message), the backend will issue an ErrorResponse

  592.     message.  If the <code class="command">COPY</code> command was issued via an extended-query

  593.     message, the backend will now discard frontend messages until a Sync

  594.     message is received, then it will issue ReadyForQuery and return to normal

  595.     processing.  If the <code class="command">COPY</code> command was issued in a simple

  596.     Query message, the rest of that message is discarded and ReadyForQuery

  597.     is issued.  In either case, any subsequent CopyData, CopyDone, or CopyFail

  598.     messages issued by the frontend will simply be dropped.

  599.    </p><p>

  600.     The backend will ignore Flush and Sync messages received during copy-in

  601.     mode.  Receipt of any other non-copy message type constitutes an error

  602.     that will abort the copy-in state as described above.  (The exception for

  603.     Flush and Sync is for the convenience of client libraries that always

  604.     send Flush or Sync after an Execute message, without checking whether

  605.     the command to be executed is a <code class="command">COPY FROM STDIN</code>.)

  606.    </p><p>

  607.     Copy-out mode (data transfer from the server) is initiated when the

  608.     backend executes a <code class="command">COPY TO STDOUT</code> SQL statement.  The backend

  609.     sends a CopyOutResponse message to the frontend, followed by

  610.     zero or more CopyData messages (always one per row), followed by CopyDone.

  611.     The backend then reverts to the command-processing mode it was

  612.     in before the <code class="command">COPY</code> started, and sends CommandComplete.

  613.     The frontend cannot abort the transfer (except by closing the connection

  614.     or issuing a Cancel request),

  615.     but it can discard unwanted CopyData and CopyDone messages.

  616.    </p><p>

  617.     In the event of a backend-detected error during copy-out mode,

  618.     the backend will issue an ErrorResponse message and revert to normal

  619.     processing.  The frontend should treat receipt of ErrorResponse as

  620.     terminating the copy-out mode.

  621.    </p><p>

  622.     It is possible for NoticeResponse and ParameterStatus messages to be

  623.     interspersed between CopyData messages; frontends must handle these cases,

  624.     and should be prepared for other asynchronous message types as well (see

  625.     <a class="xref" href="protocol-flow.html#PROTOCOL-ASYNC" title="52.2.6. Asynchronous Operations">Section 52.2.6</a>).  Otherwise, any message type other than

  626.     CopyData or CopyDone may be treated as terminating copy-out mode.

  627.    </p><p>

  628.     There is another Copy-related mode called copy-both, which allows

  629.     high-speed bulk data transfer to <span class="emphasis"><em>and</em></span> from the server.

  630.     Copy-both mode is initiated when a backend in walsender mode

  631.     executes a <code class="command">START_REPLICATION</code> statement.  The

  632.     backend sends a CopyBothResponse message to the frontend.  Both

  633.     the backend and the frontend may then send CopyData messages

  634.     until either end sends a CopyDone message. After the client

  635.     sends a CopyDone message, the connection goes from copy-both mode to

  636.     copy-out mode, and the client may not send any more CopyData messages.

  637.     Similarly, when the server sends a CopyDone message, the connection

  638.     goes into copy-in mode, and the server may not send any more CopyData

  639.     messages. After both sides have sent a CopyDone message, the copy mode

  640.     is terminated, and the backend reverts to the command-processing mode.

  641.     In the event of a backend-detected error during copy-both mode,

  642.     the backend will issue an ErrorResponse message, discard frontend messages

  643.     until a Sync message is received, and then issue ReadyForQuery and return

  644.     to normal processing.  The frontend should treat receipt of ErrorResponse

  645.     as terminating the copy in both directions; no CopyDone should be sent

  646.     in this case.  See <a class="xref" href="protocol-replication.html" title="52.4. Streaming Replication Protocol">Section 52.4</a> for more

  647.     information on the subprotocol transmitted over copy-both mode.

  648.    </p><p>

  649.     The CopyInResponse, CopyOutResponse and CopyBothResponse messages

  650.     include fields that inform the frontend of the number of columns

  651.     per row and the format codes being used for each column.  (As of

  652.     the present implementation, all columns in a given <code class="command">COPY</code>

  653.     operation will use the same format, but the message design does not

  654.     assume this.)

  655.    </p></div><div class="sect2" id="PROTOCOL-ASYNC"><div class="titlepage"><div><div><h3 class="title">52.2.6. Asynchronous Operations</h3></div></div></div><p>

  656.     There are several cases in which the backend will send messages that

  657.     are not specifically prompted by the frontend's command stream.

  658.     Frontends must be prepared to deal with these messages at any time,

  659.     even when not engaged in a query.

  660.     At minimum, one should check for these cases before beginning to

  661.     read a query response.

  662.    </p><p>

  663.     It is possible for NoticeResponse messages to be generated due to

  664.     outside activity; for example, if the database administrator commands

  665.     a <span class="quote">“<span class="quote">fast</span>”</span> database shutdown, the backend will send a NoticeResponse

  666.     indicating this fact before closing the connection.  Accordingly,

  667.     frontends should always be prepared to accept and display NoticeResponse

  668.     messages, even when the connection is nominally idle.

  669.    </p><p>

  670.     ParameterStatus messages will be generated whenever the active

  671.     value changes for any of the parameters the backend believes the

  672.     frontend should know about.  Most commonly this occurs in response

  673.     to a <code class="command">SET</code> SQL command executed by the frontend, and

  674.     this case is effectively synchronous — but it is also possible

  675.     for parameter status changes to occur because the administrator

  676.     changed a configuration file and then sent the

  677.     <span class="systemitem">SIGHUP</span> signal to the server.  Also,

  678.     if a <code class="command">SET</code> command is rolled back, an appropriate

  679.     ParameterStatus message will be generated to report the current

  680.     effective value.

  681.    </p><p>

  682.     At present there is a hard-wired set of parameters for which

  683.     ParameterStatus will be generated: they are

  684.     <code class="varname">server_version</code>,

  685.     <code class="varname">server_encoding</code>,

  686.     <code class="varname">client_encoding</code>,

  687.     <code class="varname">application_name</code>,

  688.     <code class="varname">is_superuser</code>,

  689.     <code class="varname">session_authorization</code>,

  690.     <code class="varname">DateStyle</code>,

  691.     <code class="varname">IntervalStyle</code>,

  692.     <code class="varname">TimeZone</code>,

  693.     <code class="varname">integer_datetimes</code>, and

  694.     <code class="varname">standard_conforming_strings</code>.

  695.     (<code class="varname">server_encoding</code>, <code class="varname">TimeZone</code>, and

  696.     <code class="varname">integer_datetimes</code> were not reported by releases before 8.0;

  697.     <code class="varname">standard_conforming_strings</code> was not reported by releases

  698.     before 8.1;

  699.     <code class="varname">IntervalStyle</code> was not reported by releases before 8.4;

  700.     <code class="varname">application_name</code> was not reported by releases before 9.0.)

  701.     Note that

  702.     <code class="varname">server_version</code>,

  703.     <code class="varname">server_encoding</code> and

  704.     <code class="varname">integer_datetimes</code>

  705.     are pseudo-parameters that cannot change after startup.

  706.     This set might change in the future, or even become configurable.

  707.     Accordingly, a frontend should simply ignore ParameterStatus for

  708.     parameters that it does not understand or care about.

  709.    </p><p>

  710.     If a frontend issues a <code class="command">LISTEN</code> command, then the

  711.     backend will send a NotificationResponse message (not to be

  712.     confused with NoticeResponse!)  whenever a

  713.     <code class="command">NOTIFY</code> command is executed for the same

  714.     channel name.

  715.    </p><div class="note"><h3 class="title">Note</h3><p>

  716.      At present, NotificationResponse can only be sent outside a

  717.      transaction, and thus it will not occur in the middle of a

  718.      command-response series, though it might occur just before ReadyForQuery.

  719.      It is unwise to design frontend logic that assumes that, however.

  720.      Good practice is to be able to accept NotificationResponse at any

  721.      point in the protocol.

  722.     </p></div></div><div class="sect2" id="id-1.10.5.7.9"><div class="titlepage"><div><div><h3 class="title">52.2.7. Canceling Requests in Progress</h3></div></div></div><p>

  723.     During the processing of a query, the frontend might request

  724.     cancellation of the query.  The cancel request is not sent

  725.     directly on the open connection to the backend for reasons of

  726.     implementation efficiency: we don't want to have the backend

  727.     constantly checking for new input from the frontend during query

  728.     processing.  Cancel requests should be relatively infrequent, so

  729.     we make them slightly cumbersome in order to avoid a penalty in

  730.     the normal case.

  731.    </p><p>

  732.     To issue a cancel request, the frontend opens a new connection to

  733.     the server and sends a CancelRequest message, rather than the

  734.     StartupMessage message that would ordinarily be sent across a new

  735.     connection.  The server will process this request and then close

  736.     the connection.  For security reasons, no direct reply is made to

  737.     the cancel request message.

  738.    </p><p>

  739.     A CancelRequest message will be ignored unless it contains the

  740.     same key data (PID and secret key) passed to the frontend during

  741.     connection start-up.  If the request matches the PID and secret

  742.     key for a currently executing backend, the processing of the

  743.     current query is aborted.  (In the existing implementation, this is

  744.     done by sending a special signal to the backend process that is

  745.     processing the query.)

  746.    </p><p>

  747.     The cancellation signal might or might not have any effect — for

  748.     example, if it arrives after the backend has finished processing

  749.     the query, then it will have no effect.  If the cancellation is

  750.     effective, it results in the current command being terminated

  751.     early with an error message.

  752.    </p><p>

  753.     The upshot of all this is that for reasons of both security and

  754.     efficiency, the frontend has no direct way to tell whether a

  755.     cancel request has succeeded.  It must continue to wait for the

  756.     backend to respond to the query.  Issuing a cancel simply improves

  757.     the odds that the current query will finish soon, and improves the

  758.     odds that it will fail with an error message instead of

  759.     succeeding.

  760.    </p><p>

  761.     Since the cancel request is sent across a new connection to the

  762.     server and not across the regular frontend/backend communication

  763.     link, it is possible for the cancel request to be issued by any

  764.     process, not just the frontend whose query is to be canceled.

  765.     This might provide additional flexibility when building

  766.     multiple-process applications.  It also introduces a security

  767.     risk, in that unauthorized persons might try to cancel queries.

  768.     The security risk is addressed by requiring a dynamically

  769.     generated secret key to be supplied in cancel requests.

  770.    </p></div><div class="sect2" id="id-1.10.5.7.10"><div class="titlepage"><div><div><h3 class="title">52.2.8. Termination</h3></div></div></div><p>

  771.     The normal, graceful termination procedure is that the frontend

  772.     sends a Terminate message and immediately closes the connection.

  773.     On receipt of this message, the backend closes the connection and

  774.     terminates.

  775.    </p><p>

  776.     In rare cases (such as an administrator-commanded database shutdown)

  777.     the backend might disconnect without any frontend request to do so.

  778.     In such cases the backend will attempt to send an error or notice message

  779.     giving the reason for the disconnection before it closes the connection.

  780.    </p><p>

  781.     Other termination scenarios arise from various failure cases, such as core

  782.     dump at one end or the other, loss of the communications link, loss of

  783.     message-boundary synchronization, etc.  If either frontend or backend sees

  784.     an unexpected closure of the connection, it should clean

  785.     up and terminate.  The frontend has the option of launching a new backend

  786.     by recontacting the server if it doesn't want to terminate itself.

  787.     Closing the connection is also advisable if an unrecognizable message type

  788.     is received, since this probably indicates loss of message-boundary sync.

  789.    </p><p>

  790.     For either normal or abnormal termination, any open transaction is

  791.     rolled back, not committed.  One should note however that if a

  792.     frontend disconnects while a non-<code class="command">SELECT</code> query

  793.     is being processed, the backend will probably finish the query

  794.     before noticing the disconnection.  If the query is outside any

  795.     transaction block (<code class="command">BEGIN</code> ... <code class="command">COMMIT</code>

  796.     sequence) then its results might be committed before the

  797.     disconnection is recognized.

  798.    </p></div><div class="sect2" id="id-1.10.5.7.11"><div class="titlepage"><div><div><h3 class="title">52.2.9. <acronym class="acronym">SSL</acronym> Session Encryption</h3></div></div></div><p>

  799.     If <span class="productname">PostgreSQL</span> was built with

  800.     <acronym class="acronym">SSL</acronym> support, frontend/backend communications

  801.     can be encrypted using <acronym class="acronym">SSL</acronym>.  This provides

  802.     communication security in environments where attackers might be

  803.     able to capture the session traffic. For more information on

  804.     encrypting <span class="productname">PostgreSQL</span> sessions with

  805.     <acronym class="acronym">SSL</acronym>, see <a class="xref" href="ssl-tcp.html" title="18.9. Secure TCP/IP Connections with SSL">Section 18.9</a>.

  806.    </p><p>

  807.     To initiate an <acronym class="acronym">SSL</acronym>-encrypted connection, the

  808.     frontend initially sends an SSLRequest message rather than a

  809.     StartupMessage.  The server then responds with a single byte

  810.     containing <code class="literal">S</code> or <code class="literal">N</code>, indicating that it is

  811.     willing or unwilling to perform <acronym class="acronym">SSL</acronym>,

  812.     respectively.  The frontend might close the connection at this point

  813.     if it is dissatisfied with the response.  To continue after

  814.     <code class="literal">S</code>, perform an <acronym class="acronym">SSL</acronym> startup handshake

  815.     (not described here, part of the <acronym class="acronym">SSL</acronym>

  816.     specification) with the server.  If this is successful, continue

  817.     with sending the usual StartupMessage.  In this case the

  818.     StartupMessage and all subsequent data will be

  819.     <acronym class="acronym">SSL</acronym>-encrypted.  To continue after

  820.     <code class="literal">N</code>, send the usual StartupMessage and proceed without

  821.     encryption.

  822.    </p><p>

  823.     The frontend should also be prepared to handle an ErrorMessage

  824.     response to SSLRequest from the server.  This would only occur if

  825.     the server predates the addition of <acronym class="acronym">SSL</acronym> support

  826.     to <span class="productname">PostgreSQL</span>.  (Such servers are now very ancient,

  827.     and likely do not exist in the wild anymore.)

  828.     In this case the connection must

  829.     be closed, but the frontend might choose to open a fresh connection

  830.     and proceed without requesting <acronym class="acronym">SSL</acronym>.

  831.    </p><p>

  832.     An initial SSLRequest can also be used in a connection that is being

  833.     opened to send a CancelRequest message.

  834.    </p><p>

  835.     While the protocol itself does not provide a way for the server to

  836.     force <acronym class="acronym">SSL</acronym> encryption, the administrator can

  837.     configure the server to reject unencrypted sessions as a byproduct

  838.     of authentication checking.

  839.    </p></div><div class="sect2" id="id-1.10.5.7.12"><div class="titlepage"><div><div><h3 class="title">52.2.10. <acronym class="acronym">GSSAPI</acronym> Session Encryption</h3></div></div></div><p>

  840.     If <span class="productname">PostgreSQL</span> was built with

  841.     <acronym class="acronym">GSSAPI</acronym> support, frontend/backend communications

  842.     can be encrypted using <acronym class="acronym">GSSAPI</acronym>.  This provides

  843.     communication security in environments where attackers might be

  844.     able to capture the session traffic. For more information on

  845.     encrypting <span class="productname">PostgreSQL</span> sessions with

  846.     <acronym class="acronym">GSSAPI</acronym>, see <a class="xref" href="gssapi-enc.html" title="18.10. Secure TCP/IP Connections with GSSAPI Encryption">Section 18.10</a>.

  847.    </p><p>

  848.     To initiate a <acronym class="acronym">GSSAPI</acronym>-encrypted connection, the

  849.     frontend initially sends a GSSENCRequest message rather than a

  850.     StartupMessage.  The server then responds with a single byte

  851.     containing <code class="literal">G</code> or <code class="literal">N</code>, indicating that it

  852.     is willing or unwilling to perform <acronym class="acronym">GSSAPI</acronym> encryption,

  853.     respectively.  The frontend might close the connection at this point

  854.     if it is dissatisfied with the response.  To continue after

  855.     <code class="literal">G</code>, using the GSSAPI C bindings as discussed in RFC2744

  856.     or equivalent, perform a <acronym class="acronym">GSSAPI</acronym> initialization by

  857.     calling <code class="function">gss_init_sec_context()</code> in a loop and sending

  858.     the result to the server, starting with an empty input and then with each

  859.     result from the server, until it returns no output.  When sending the

  860.     results of <code class="function">gss_init_sec_context()</code> to the server,

  861.     prepend the length of the message as a four byte integer in network byte

  862.     order.  If this is successful, then use <code class="function">gss_wrap()</code> to

  863.     encrypt the usual StartupMessage and all subsequent data, prepending the

  864.     length of the result from <code class="function">gss_wrap()</code> as a four byte

  865.     integer in network byte order to the actual encrypted payload.  Note that

  866.     the server will only accept encrypted packets from the client which are less

  867.     than 16kB; <code class="function">gss_wrap_size_limit()</code> should be used by the

  868.     client to determine the size of the unencrypted message which will fit

  869.     within this limit and larger messages should be broken up into multiple

  870.     <code class="function">gss_wrap()</code> calls.  Typical segments are 8kB of

  871.     unencrypted data, resulting in encrypted packets of slightly larger than 8kB

  872.     but well within the 16kB maximum.  The server can be expected to not send

  873.     encrypted packets of larger than 16kB to the client.  To continue after

  874.     <code class="literal">N</code>, send the usual StartupMessage and proceed without

  875.     encryption.

  876.    </p><p>

  877.     The frontend should also be prepared to handle an ErrorMessage

  878.     response to GSSENCRequest from the server.  This would only occur if

  879.     the server predates the addition of <acronym class="acronym">GSSAPI</acronym> encryption

  880.     support to <span class="productname">PostgreSQL</span>.  In this case the

  881.     connection must be closed, but the frontend might choose to open a fresh

  882.     connection and proceed without requesting <acronym class="acronym">GSSAPI</acronym>

  883.     encryption.  Given the length limits specified above, the ErrorMessage can

  884.     not be confused with a proper response from the server with an appropriate

  885.     length.

  886.    </p><p>

  887.     An initial GSSENCRequest can also be used in a connection that is being

  888.     opened to send a CancelRequest message.

  889.    </p><p>

  890.     While the protocol itself does not provide a way for the server to

  891.     force <acronym class="acronym">GSSAPI</acronym> encryption, the administrator can

  892.     configure the server to reject unencrypted sessions as a byproduct

  893.     of authentication checking.

  894.    </p></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="protocol-overview.html">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="protocol.html">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="sasl-authentication.html">Next</a></td></tr><tr><td width="40%" align="left" valign="top">52.1. Overview </td><td width="20%" align="center"><a accesskey="h" href="index.html">Home</a></td><td width="40%" align="right" valign="top"> 52.3. SASL Authentication</td></tr></table></div></body></html>
上海开阖软件有限公司 沪ICP备12045867号-1