3 <!ENTITY project SYSTEM "project.xml">
5 <document url="ajpv13a.html">
9 Licensed to the Apache Software Foundation (ASF) under one or more
10 contributor license agreements. See the NOTICE file distributed with
11 this work for additional information regarding copyright ownership.
12 The ASF licenses this file to You under the Apache License, Version 2.0
13 (the "License"); you may not use this file except in compliance with
14 the License. You may obtain a copy of the License at
16 http://www.apache.org/licenses/LICENSE-2.0
18 Unless required by applicable law or agreed to in writing, software
19 distributed under the License is distributed on an "AS IS" BASIS,
20 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
21 See the License for the specific language governing permissions and
22 limitations under the License.
26 <author email="danmil@shore.net">danmil@shore.net</author>
27 <author email="jfrederic.clere@fujitsu-siemens.com">Jean-Frederic Clere</author>
28 <date>$Date: 2007-06-09 22:38:06 +0200 (Sat, 09 Jun 2007) $</date>
31 <section name="Intro">
34 The original document was written by
35 Dan Milstein, <author email="danmil@shore.net">danmil@shore.net</author>
36 on December 2000. The present document is generated out of an xml file
37 to allow a more easy integration in the Tomcat documentation.
42 This describes the Apache JServ Protocol version 1.3 (hereafter
43 <b>ajp13</b>). There is, apparently, no current documentation of how the
44 protocol works. This document is an attempt to remedy that, in order to
45 make life easier for maintainers of JK, and for anyone who wants to
46 port the protocol somewhere (into jakarta 4.x, for example).
51 <section name="author">
54 I am not one of the designers of this protocol -- I believe that Gal
55 Shachor was the original designer. Everything in this document is derived
56 from the actual implementation I found in the tomcat 3.x code. I hope it
57 is useful, but I can't make any grand claims to perfect accuracy. I also
58 don't know why certain design decisions were made. Where I was able, I've
59 offered some possible justifications for certain choices, but those are
60 only my guesses. In general, the C code which Shachor wrote is very clean
61 and comprehensible (if almost totally undocumented). I've cleaned up the
62 Java code, and I think it's reasonably readable.
66 <section name="Design Goals">
69 According to email from Gal Shachor to the jakarta-dev mailing list,
70 the original goals of <b>JK</b> (and thus <b>ajp13</b>) were to extend
71 <b>mod_jserv</b> and <b>ajp12</b> by (I am only including the goals which
72 relate to communication between the web server and the servlet container):
75 <li> Increasing performance (speed, specifically). </li>
77 <li> Adding support for SSL, so that <code>isSecure()</code> and
78 <code>getScheme()</code> will function correctly within the servlet
79 container. The client certificates and cipher suite will be
80 available to servlets as request attributes. </li>
86 <section name="Overview of the protocol">
89 The <b>ajp13</b> protocol is packet-oriented. A binary format was
90 presumably chosen over the more readable plain text for reasons of
91 performance. The web server communicates with the servlet container over
92 TCP connections. To cut down on the expensive process of socket creation,
93 the web server will attempt to maintain persistent TCP connections to the
94 servlet container, and to reuse a connection for multiple request/response
97 Once a connection is assigned to a particular request, it will not be
98 used for any others until the request-handling cycle has terminated. In
99 other words, requests are not multiplexed over connections. This makes
100 for much simpler code at either end of the connection, although it does
101 cause more connections to be open at once.
103 Once the web server has opened a connection to the servlet container,
104 the connection can be in one of the following states:
107 <li> Idle <br/> No request is being handled over this connection. </li>
108 <li> Assigned <br/> The connecton is handling a specific request.</li>
112 Once a connection is assigned to handle a particular request, the basic
113 request informaton (e.g. HTTP headers, etc) is sent over the connection in
114 a highly condensed form (e.g. common strings are encoded as integers).
115 Details of that format are below in Request Packet Structure. If there is a
116 body to the request (content-length > 0), that is sent in a separate
117 packet immediately after.
119 At this point, the servlet container is presumably ready to start
120 processing the request. As it does so, it can send the
121 following messages back to the web server:
124 <li>SEND_HEADERS <br/>Send a set of headers back to the browser.</li>
126 <li>SEND_BODY_CHUNK <br/>Send a chunk of body data back to the browser.</li>
128 <li>GET_BODY_CHUNK <br/>Get further data from the request if it hasn't all
129 been transferred yet. This is necessary because the packets have a fixed
130 maximum size and arbitrary amounts of data can be included the body of a
131 request (for uploaded files, for example). (Note: this is unrelated to
132 HTTP chunked tranfer).</li>
134 <li>END_RESPONSE <br/> Finish the request-handling cycle.</li>
138 Each message is accompanied by a differently formatted packet of data. See
139 Response Packet Structures below for details.
143 <section name="Basic Packet Structure">
146 There is a bit of an XDR heritage to this protocol, but it differs in
147 lots of ways (no 4 byte alignment, for example).
149 Byte order: I am not clear about the endian-ness of the individual
150 bytes. I'm guessing the bytes are little-endian, because that's what XDR
151 specifies, and I'm guessing that sys/socket library is magically making
152 that so (on the C side). If anyone with a better knowledge of socket calls
153 can step in, that would be great.
155 There are four data types in the protocol: bytes, booleans, integers and
160 <dd>A single byte.</dd>
162 <dt><b>Boolean</b></dt>
163 <dd>A single byte, 1 = true, 0 = false. Using other non-zero values as
164 true (i.e. C-style) may work in some places, but it won't in
167 <dt><b>Integer</b></dt>
168 <dd>A number in the range of 0 to 2^16 (32768). Stored in 2 bytes with
169 the high-order byte first.</dd>
171 <dt><b>String</b></dt>
172 <dd>A variable-sized string (length bounded by 2^16). Encoded with the
173 length packed into two bytes first, followed by the string (including the
174 terminating '\0'). Note that the encoded length does <b>not</b> include
175 the trailing '\0' -- it is like <code>strlen</code>. This is a touch
176 confusing on the Java side, which is littered with odd autoincrement
177 statements to skip over these terminators. I believe the reason this was
178 done was to allow the C code to be extra efficient when reading strings
179 which the servlet container is sending back -- with the terminating \0
180 character, the C code can pass around references into a single buffer,
181 without copying. If the \0 was missing, the C code would have to copy
182 things out in order to get its notion of a string. Note a size of -1
183 (65535) indicates a null string and no data follow the length in this
188 <subsection name="Packet Size">
190 According to much of the code, the max packet
191 size is 8 * 1024 bytes (8K). The actual length of the packet is encoded in the
196 <subsection name="Packet Headers">
198 Packets sent from the server to the container begin with
199 <code>0x1234</code>. Packets sent from the container to the server begin
200 with <code>AB</code> (that's the ASCII code for A followed by the ASCII
201 code for B). After those first two bytes, there is an integer (encoded as
202 above) with the length of the payload. Although this might suggest that
203 the maximum payload could be as large as 2^16, in fact, the code sets the
209 <th colspan="6">Packet Format (Server->Container)</th>
225 <td colspan="2">Data Length (n)</td>
232 <th colspan="6"><b>Packet Format (Container->Server)</b></th>
248 <td colspan="2">Data Length (n)</td>
254 <A NAME="prefix-codes"></A> For most packets, the first byte of the
255 payload encodes the type of message. The exception is for request body
256 packets sent from the server to the container -- they are sent with a
257 standard packet header (0x1234 and then length of the packet), but without
258 any prefix code after that (this seems like a mistake to me).
260 The web server can send the following messages to the servlet container:
265 <th>Type of Packet</th>
270 <td>Forward Request</td>
271 <td>Begin the request-processing cycle with the following data</td>
276 <td>The web server asks the container to shut itself down.</td>
281 <td>The web server asks the container to take control (secure login phase).</td>
286 <td>The web server asks the container to respond quickly with a CPong.</td>
291 <td>Size (2 bytes) and corresponding body data.</td>
297 basic security, the container will only actually do the <code>Shutdown</code> if the
298 request comes from the same machine on which it's hosted.
301 The first <code>Data</code> packet is send immediatly after the <code>Forward Request</code> by the web server.
304 <p>The servlet container can send the following types of messages to the web
309 <th>Type of Packet</th>
314 <td>Send Body Chunk</td>
315 <td>Send a chunk of the body from the servlet container to the web
316 server (and presumably, onto the browser). </td>
320 <td>Send Headers</td>
321 <td>Send the response headers from the servlet container to the web
322 server (and presumably, onto the browser).</td>
326 <td>End Response</td>
327 <td>Marks the end of the response (and thus the request-handling cycle).</td>
331 <td>Get Body Chunk</td>
332 <td>Get further data from the request if it hasn't all been transferred
338 <td>The reply to a CPing request</td>
343 Each of the above messages has a different internal structure, detailed below.
348 <section name="Request Packet Structure">
351 For messages from the server to the container of type "Forward Request":
354 AJP13_FORWARD_REQUEST :=
355 prefix_code (byte) 0x02 = JK_AJP13_FORWARD_REQUEST
362 server_port (integer)
364 num_headers (integer)
365 request_headers *(req_header_name req_header_value)
366 attributes *(attribut_name attribute_value)
367 request_terminator (byte) OxFF
370 The <code>request_headers</code> have the following structure:
374 sc_req_header_name | (string) [see below for how this is parsed]
376 sc_req_header_name := 0xA0xx (integer)
378 req_header_value := (string)
382 The <code>attributes</code> are optional and have the following structure:
385 attribute_name := sc_a_name | (sc_a_req_attribute string)
387 attribute_value := (string)
391 Not that the all-important header is "content-length', because it
392 determines whether or not the container looks for another packet
395 Detailed description of the elements of Forward Request.
397 <subsection name="request_prefix">
399 For all requests, this will be 2.
400 See above for details on other <A HREF="#prefix-codes">prefix codes</A>.
404 <subsection name="method">
406 The HTTP method, encoded as a single byte:
411 <tr><th>Command Name</th><th>Code</th></tr>
412 <tr><td>OPTIONS</td><td>1</td></tr>
413 <tr><td>GET</td><td>2</td></tr>
414 <tr><td>HEAD</td><td>3</td></tr>
415 <tr><td>POST</td><td>4</td></tr>
416 <tr><td>PUT</td><td>5</td></tr>
417 <tr><td>DELETE</td><td>6</td></tr>
418 <tr><td>TRACE</td><td>7</td></tr>
419 <tr><td>PROPFIND</td><td>8</td></tr>
420 <tr><td>PROPPATCH</td><td>9</td></tr>
421 <tr><td>MKCOL</td><td>10</td></tr>
422 <tr><td>COPY</td><td>11</td></tr>
423 <tr><td>MOVE</td><td>12</td></tr>
424 <tr><td>LOCK</td><td>13</td></tr>
425 <tr><td>UNLOCK</td><td>14</td></tr>
426 <tr><td>ACL</td><td>15</td></tr>
427 <tr><td>REPORT</td><td>16</td></tr>
428 <tr><td>VERSION-CONTROL</td><td>17</td></tr>
429 <tr><td>CHECKIN</td><td>18</td></tr>
430 <tr><td>CHECKOUT</td><td>19</td></tr>
431 <tr><td>UNCHECKOUT</td><td>20</td></tr>
432 <tr><td>SEARCH</td><td>21</td></tr>
433 <tr><td>MKWORKSPACE</td><td>22</td></tr>
434 <tr><td>UPDATE</td><td>23</td></tr>
435 <tr><td>LABEL</td><td>24</td></tr>
436 <tr><td>MERGE</td><td>25</td></tr>
437 <tr><td>BASELINE_CONTROL</td><td>26</td></tr>
438 <tr><td>MKACTIVITY</td><td>27</td></tr>
442 <p>Later version of ajp13, when used with mod_jk2, will transport
443 additional methods, even if they are not in this list.
448 <subsection name="protocol, req_uri, remote_addr, remote_host, server_name, server_port, is_ssl">
450 These are all fairly self-explanatory. Each of these is required, and
451 will be sent for every request.
455 <subsection name="Headers">
457 The structure of <code>request_headers</code> is the following:
458 First, the number of headers <code>num_headers</code> is encoded.
459 Then, a series of header name <code>req_header_name</code> / value
460 <code>req_header_value</code> pairs follows.
461 Common header names are encoded as integers,
462 to save space. If the header name is not in the list of basic headers,
463 it is encoded normally (as a string, with prefixed length). The list of
464 common headers <code>sc_req_header_name</code>and their codes
465 is as follows (all are case-sensitive):
468 <tr><th>Name</th><th>Code value</th><th>Code name</th></tr>
469 <tr><td>accept</td><td>0xA001</td><td>SC_REQ_ACCEPT</td></tr>
470 <tr><td>accept-charset</td><td>0xA002</td><td>SC_REQ_ACCEPT_CHARSET</td></tr>
471 <tr><td>accept-encoding</td><td>0xA003</td><td>SC_REQ_ACCEPT_ENCODING</td></tr>
472 <tr><td>accept-language</td><td>0xA004</td><td>SC_REQ_ACCEPT_LANGUAGE</td></tr>
473 <tr><td>authorization</td><td>0xA005</td><td>SC_REQ_AUTHORIZATION</td></tr>
474 <tr><td>connection</td><td>0xA006</td><td>SC_REQ_CONNECTION</td></tr>
475 <tr><td>content-type</td><td>0xA007</td><td>SC_REQ_CONTENT_TYPE</td></tr>
476 <tr><td>content-length</td><td>0xA008</td><td>SC_REQ_CONTENT_LENGTH</td></tr>
477 <tr><td>cookie</td><td>0xA009</td><td>SC_REQ_COOKIE</td></tr>
478 <tr><td>cookie2</td><td>0xA00A</td><td>SC_REQ_COOKIE2</td></tr>
479 <tr><td>host</td><td>0xA00B</td><td>SC_REQ_HOST</td></tr>
480 <tr><td>pragma</td><td>0xA00C</td><td>SC_REQ_PRAGMA</td></tr>
481 <tr><td>referer</td><td>0xA00D</td><td>SC_REQ_REFERER</td></tr>
482 <tr><td>user-agent</td><td>0xA00E</td><td>SC_REQ_USER_AGENT</td></tr>
485 The Java code that reads this grabs the first two-byte integer and if
486 it sees an <code>'0xA0'</code> in the most significant
487 byte, it uses the integer in the second byte as an index into an array of
488 header names. If the first byte is not '0xA0', it assumes that the
489 two-byte integer is the length of a string, which is then read in.
491 This works on the assumption that no header names will have length
492 greater than 0x9999 (==0xA000 - 1), which is perfectly reasonable, though
493 somewhat arbitrary. (If you, like me, started to think about the cookie
494 spec here, and about how long headers can get, fear not -- this limit is
495 on header <b>names</b> not header <b>values</b>. It seems unlikely that
496 unmanageably huge header names will be showing up in the HTTP spec any time
499 <b>Note:</b> The <code>content-length</code> header is extremely
500 important. If it is present and non-zero, the container assumes that
501 the request has a body (a POST request, for example), and immediately
502 reads a separate packet off the input stream to get that body.
506 <subsection name="Attributes">
509 The attributes prefixed with a <code>?</code>
510 (e.g. <code>?context</code>) are all optional. For each, there is a
511 single byte code to indicate the type of attribute, and then a string to
512 give its value. They can be sent in any order (thogh the C code always
513 sends them in the order listed below). A special terminating code is
514 sent to signal the end of the list of optional attributes. The list of
519 <tr><th>Information</th><th>Code Value</th><th>Note</th></tr>
520 <tr><td>?context</td><td>0x01</td><td>Not currently implemented</td></tr>
521 <tr><td>?servlet_path</td><td>0x02</td><td>Not currently implemented</td></tr>
522 <tr><td>?remote_user</td><td>0x03</td><td></td></tr>
523 <tr><td>?auth_type</td><td>0x04</td><td></td></tr>
524 <tr><td>?query_string</td><td>0x05</td><td></td></tr>
525 <tr><td>?route</td><td>0x06</td><td></td></tr>
526 <tr><td>?ssl_cert</td><td>0x07</td><td></td></tr>
527 <tr><td>?ssl_cipher</td><td>0x08</td><td></td></tr>
528 <tr><td>?ssl_session</td><td>0x09</td><td></td></tr>
529 <tr><td>?req_attribute</td><td>0x0A</td><td>Name (the name of the attribut follows)</td></tr>
530 <tr><td>?ssl_key_size</td><td>0x0B</td><td></td></tr>
531 <tr><td>?secret</td><td>0x0C</td><td></td></tr>
532 <tr><td>?stored_method</td><td>0x0D</td><td></td></tr>
533 <tr><td>are_done</td><td>0xFF</td><td>request_terminator</td></tr>
538 The <code>context</code> and <code>servlet_path</code> are not currently
539 set by the C code, and most of the Java code completely ignores whatever
540 is sent over for those fields (and some of it will actually break if a
541 string is sent along after one of those codes). I don't know if this is
542 a bug or an unimplemented feature or just vestigial code, but it's
543 missing from both sides of the connection.
545 The <code>remote_user</code> and <code>auth_type</code> presumably refer
546 to HTTP-level authentication, and communicate the remote user's username
547 and the type of authentication used to establish their identity (e.g. Basic,
548 Digest). I'm not clear on why the password isn't also sent, but I don't
549 know HTTP authentication inside and out.
551 The <code>query_string</code>, <code>ssl_cert</code>,
552 <code>ssl_cipher</code>, and <code>ssl_session</code> refer to the
553 corresponding pieces of HTTP and HTTPS.
555 The <code>route</code>, as I understand it, is used to support sticky
556 sessions -- associating a user's sesson with a particular Tomcat instance
557 in the presence of multiple, load-balancing servers. I don't know the
560 Beyond this list of basic attributes, any number of other attributes can
561 be sent via the <code>req_attribute</code> code (0x0A). A pair of strings
562 to represent the attribute name and value are sent immediately after each
563 instance of that code. Environment values are passed in via this method.
565 Finally, after all the attributes have been sent, the attribute terminator,
566 0xFF, is sent. This signals both the end of the list of attributes and
567 also then end of the Request Packet.
573 <section name="Response Packet Structure">
576 For messages which the container can send back to the server.
579 AJP13_SEND_BODY_CHUNK :=
581 chunk_length (integer)
585 AJP13_SEND_HEADERS :=
587 http_status_code (integer)
588 http_status_msg (string)
589 num_headers (integer)
590 response_headers *(res_header_name header_value)
593 sc_res_header_name | (string) [see below for how this is parsed]
595 sc_res_header_name := 0xA0 (byte)
597 header_value := (string)
599 AJP13_END_RESPONSE :=
604 AJP13_GET_BODY_CHUNK :=
606 requested_length (integer)
614 <subsection name="Send Body Chunk">
616 The chunk is basically binary data, and is sent directly back to the browser.
620 <subsection name="Send Headers">
622 The status code and message are the usual HTTP things (e.g. "200" and "OK").
623 The response header names are encoded the same way the request header names are.
624 See <A HREF="#header_encoding">above</A> for details about how the the
625 codes are distinguished from the strings. The codes for common headers are:
630 <tr><th>Name</th><th>Code value</th></tr>
631 <tr><td>Content-Type</td><td>0xA001</td></tr>
632 <tr><td>Content-Language</td><td>0xA002</td></tr>
633 <tr><td>Content-Length</td><td>0xA003</td></tr>
634 <tr><td>Date</td><td>0xA004</td></tr>
635 <tr><td>Last-Modified</td><td>0xA005</td></tr>
636 <tr><td>Location</td><td>0xA006</td></tr>
637 <tr><td>Set-Cookie</td><td>0xA007</td></tr>
638 <tr><td>Set-Cookie2</td><td>0xA008</td></tr>
639 <tr><td>Servlet-Engine</td><td>0xA009</td></tr>
640 <tr><td>Status</td><td>0xA00A</td></tr>
641 <tr><td>WWW-Authenticate</td><td>0xA00B</td></tr>
647 After the code or the string header name, the header value is immediately
653 <subsection name="End Response">
655 Signals the end of this request-handling cycle. If the
656 <code>reuse</code> flag is true (==1), this TCP connection can now be used to
657 handle new incoming requests. If <code>reuse</code> is false (anything
658 other than 1 in the actual C code), the connection should be closed.
662 <subsection name="Get Body Chunk">
664 The container asks for more data from the request (If the body was
665 too large to fit in the first packet sent over or when the request is
667 The server will send a body packet back with an amount of data which is
668 the minimum of the <code>request_length</code>,
669 the maximum send body size (8186 (8 Kbytes - 6)), and the
670 number of bytes actually left to send from the request body.
672 If there is no more data in the body (i.e. the servlet container is
673 trying to read past the end of the body), the server will send back an
674 "empty" packet, which is a body packet with a payload length of 0.
675 (0x12,0x34,0x00,0x00)
680 <section name="Questions I Have">
682 <p> What happens if the request headers > max packet size? There is no
683 provision to send a second packet of request headers in case there are more
684 than 8K (I think this is correctly handled for response headers, though I'm
685 not certain). I don't know if there is a way to get more than 8K worth of
686 data into that initial set of request headers, but I'll bet there is
687 (combine long cookies with long ssl information and a lot of environment
688 variables, and you should hit 8K easily). I think the connector would just
689 fail before trying to send any headers in this case, but I'm not certain.</p>
691 <p> What about authentication? There doesn't seem to be any authentication
692 of the connection between the web server and the container. This strikes
693 me as potentially dangerous.</p>