1 Git Wire Protocol, Version 2
2 ============================
4 This document presents a specification for a version 2 of Git's wire
5 protocol. Protocol v2 will improve upon v1 in the following ways:
7 * Instead of multiple service names, multiple commands will be
8 supported by a single service
9 * Easily extendable as capabilities are moved into their own section
10 of the protocol, no longer being hidden behind a NUL byte and
11 limited by the size of a pkt-line
12 * Separate out other information hidden behind NUL bytes (e.g. agent
13 string as a capability and symrefs can be requested using 'ls-refs')
14 * Reference advertisement will be omitted unless explicitly requested
15 * ls-refs command to explicitly request some refs
16 * Designed with http and stateless-rpc in mind. With clear flush
17 semantics the http remote helper can simply act as a proxy
19 In protocol v2 communication is command oriented. When first contacting a
20 server a list of capabilities will advertised. Some of these capabilities
21 will be commands which a client can request be executed. Once a command
22 has completed, a client can reuse the connection and request that other
28 All communication is done using packet-line framing, just as in v1. See
29 `Documentation/technical/pack-protocol.txt` and
30 `Documentation/technical/protocol-common.txt` for more information.
32 In protocol v2 these special packets will have the following semantics:
34 * '0000' Flush Packet (flush-pkt) - indicates the end of a message
35 * '0001' Delimiter Packet (delim-pkt) - separates sections of a message
36 * '0002' Response End Packet (response-end-pkt) - indicates the end of a
37 response for stateless connections
39 Initial Client Request
40 ----------------------
42 In general a client can request to speak protocol v2 by sending
43 `version=2` through the respective side-channel for the transport being
44 used which inevitably sets `GIT_PROTOCOL`. More information can be
45 found in `pack-protocol.txt` and `http-protocol.txt`. In all cases the
46 response from the server is the capability advertisement.
51 When using the git:// transport, you can request to use protocol v2 by
52 sending "version=2" as an extra parameter:
54 003egit-upload-pack /project.git\0host=myserver.com\0\0version=2\0
56 SSH and File Transport
57 ~~~~~~~~~~~~~~~~~~~~~~
59 When using either the ssh:// or file:// transport, the GIT_PROTOCOL
60 environment variable must be set explicitly to include "version=2".
65 When using the http:// or https:// transport a client makes a "smart"
66 info/refs request as described in `http-protocol.txt` and requests that
67 v2 be used by supplying "version=2" in the `Git-Protocol` header.
69 C: GET $GIT_URL/info/refs?service=git-upload-pack HTTP/1.0
70 C: Git-Protocol: version=2
72 A v2 server would reply:
79 S: <capability-advertisement>
81 Subsequent requests are then made directly to the service
82 `$GIT_URL/git-upload-pack`. (This works the same for git-receive-pack).
84 Capability Advertisement
85 ------------------------
87 A server which decides to communicate (based on a request from a client)
88 using protocol version 2, notifies the client by sending a version string
89 in its initial response followed by an advertisement of its capabilities.
90 Each capability is a key with an optional value. Clients must ignore all
91 unknown keys. Semantics of unknown values are left to the definition of
92 each key. Some capabilities will describe commands which can be requested
93 to be executed by the client.
95 capability-advertisement = protocol-version
99 protocol-version = PKT-LINE("version 2" LF)
100 capability-list = *capability
101 capability = PKT-LINE(key[=value] LF)
103 key = 1*(ALPHA | DIGIT | "-_")
104 value = 1*(ALPHA | DIGIT | " -_.,?\/{}[]()<>!@#$%^&*+=:;")
109 After receiving the capability advertisement, a client can then issue a
110 request to select the command it wants with any particular capabilities
111 or arguments. There is then an optional section where the client can
112 provide any command specific parameters or queries. Only a single
113 command can be requested at a time.
115 request = empty-request | command-request
116 empty-request = flush-pkt
117 command-request = command
121 command = PKT-LINE("command=" key LF)
122 command-args = delim-pkt
123 *command-specific-arg
125 command-specific-args are packet line framed arguments defined by
126 each individual command.
128 The server will then check to ensure that the client's request is
129 comprised of a valid command as well as valid capabilities which were
130 advertised. If the request is valid the server will then execute the
131 command. A server MUST wait till it has received the client's entire
132 request before issuing a response. The format of the response is
133 determined by the command being executed, but in all cases a flush-pkt
134 indicates the end of the response.
136 When a command has finished, and the client has received the entire
137 response from the server, a client can either request that another
138 command be executed or can terminate the connection. A client may
139 optionally send an empty request consisting of just a flush-pkt to
140 indicate that no more requests will be made.
145 There are two different types of capabilities: normal capabilities,
146 which can be used to convey information or alter the behavior of a
147 request, and commands, which are the core actions that a client wants to
148 perform (fetch, push, etc).
150 Protocol version 2 is stateless by default. This means that all commands
151 must only last a single round and be stateless from the perspective of the
152 server side, unless the client has requested a capability indicating that
153 state should be maintained by the server. Clients MUST NOT require state
154 management on the server side in order to function correctly. This
155 permits simple round-robin load-balancing on the server side, without
156 needing to worry about state management.
161 The server can advertise the `agent` capability with a value `X` (in the
162 form `agent=X`) to notify the client that the server is running version
163 `X`. The client may optionally send its own agent string by including
164 the `agent` capability with a value `Y` (in the form `agent=Y`) in its
165 request to the server (but it MUST NOT do so if the server did not
166 advertise the agent capability). The `X` and `Y` strings may contain any
167 printable ASCII characters except space (i.e., the byte range 32 < x <
168 127), and are typically of the form "package/version" (e.g.,
169 "git/1.8.3.1"). The agent strings are purely informative for statistics
170 and debugging purposes, and MUST NOT be used to programmatically assume
171 the presence or absence of particular features.
176 `ls-refs` is the command used to request a reference advertisement in v2.
177 Unlike the current reference advertisement, ls-refs takes in arguments
178 which can be used to limit the refs sent from the server.
180 Additional features not supported in the base command will be advertised
181 as the value of the command in the capability advertisement in the form
182 of a space separated list of features: "<command>=<feature 1> <feature 2>"
184 ls-refs takes in the following arguments:
187 In addition to the object pointed by it, show the underlying ref
188 pointed by it when showing a symbolic ref.
192 When specified, only references having a prefix matching one of
193 the provided prefixes are displayed.
195 If the 'unborn' feature is advertised the following argument can be
196 included in the client's request.
199 The server will send information about HEAD even if it is a symref
200 pointing to an unborn branch in the form "unborn HEAD
201 symref-target:<target>".
203 The output of ls-refs is as follows:
207 obj-id-or-unborn = (obj-id | "unborn")
208 ref = PKT-LINE(obj-id-or-unborn SP refname *(SP ref-attribute) LF)
209 ref-attribute = (symref | peeled)
210 symref = "symref-target:" symref-target
211 peeled = "peeled:" obj-id
216 `fetch` is the command used to fetch a packfile in v2. It can be looked
217 at as a modified version of the v1 fetch where the ref-advertisement is
218 stripped out (since the `ls-refs` command fills that role) and the
219 message format is tweaked to eliminate redundancies and permit easy
220 addition of future extensions.
222 Additional features not supported in the base command will be advertised
223 as the value of the command in the capability advertisement in the form
224 of a space separated list of features: "<command>=<feature 1> <feature 2>"
226 A `fetch` request can take the following arguments:
229 Indicates to the server an object which the client wants to
230 retrieve. Wants can be anything and are not limited to
234 Indicates to the server an object which the client has locally.
235 This allows the server to make a packfile which only contains
236 the objects that the client needs. Multiple 'have' lines can be
240 Indicates to the server that negotiation should terminate (or
241 not even begin if performing a clone) and that the server should
242 use the information supplied in the request to construct the
246 Request that a thin pack be sent, which is a pack with deltas
247 which reference base objects not contained within the pack (but
248 are known to exist at the receiving end). This can reduce the
249 network traffic significantly, but it requires the receiving end
250 to know how to "thicken" these packs by adding the missing bases
254 Request that progress information that would normally be sent on
255 side-band channel 2, during the packfile transfer, should not be
256 sent. However, the side-band channel 3 is still used for error
260 Request that annotated tags should be sent if the objects they
261 point to are being sent.
264 Indicate that the client understands PACKv2 with delta referring
265 to its base by position in pack rather than by an oid. That is,
266 they can read OBJ_OFS_DELTA (aka type 6) in a packfile.
268 If the 'shallow' feature is advertised the following arguments can be
269 included in the clients request as well as the potential addition of the
270 'shallow-info' section in the server's response as explained below.
273 A client must notify the server of all commits for which it only
274 has shallow copies (meaning that it doesn't have the parents of
275 a commit) by supplying a 'shallow <oid>' line for each such
276 object so that the server is aware of the limitations of the
277 client's history. This is so that the server is aware that the
278 client may not have all objects reachable from such commits.
281 Requests that the fetch/clone should be shallow having a commit
282 depth of <depth> relative to the remote side.
285 Requests that the semantics of the "deepen" command be changed
286 to indicate that the depth requested is relative to the client's
287 current shallow boundary, instead of relative to the requested
290 deepen-since <timestamp>
291 Requests that the shallow clone/fetch should be cut at a
292 specific time, instead of depth. Internally it's equivalent to
293 doing "git rev-list --max-age=<timestamp>". Cannot be used with
297 Requests that the shallow clone/fetch should be cut at a
298 specific revision specified by '<rev>', instead of a depth.
299 Internally it's equivalent of doing "git rev-list --not <rev>".
300 Cannot be used with "deepen", but can be used with
303 If the 'filter' feature is advertised, the following argument can be
304 included in the client's request:
307 Request that various objects from the packfile be omitted
308 using one of several filtering techniques. These are intended
309 for use with partial clone and partial fetch operations. See
310 `rev-list` for possible "filter-spec" values. When communicating
311 with other processes, senders SHOULD translate scaled integers
312 (e.g. "1k") into a fully-expanded form (e.g. "1024") to aid
313 interoperability with older receivers that may not understand
314 newly-invented scaling suffixes. However, receivers SHOULD
315 accept the following suffixes: 'k', 'm', and 'g' for 1024,
316 1048576, and 1073741824, respectively.
318 If the 'ref-in-want' feature is advertised, the following argument can
319 be included in the client's request as well as the potential addition of
320 the 'wanted-refs' section in the server's response as explained below.
323 Indicates to the server that the client wants to retrieve a
324 particular ref, where <ref> is the full name of a ref on the
327 If the 'sideband-all' feature is advertised, the following argument can be
328 included in the client's request:
331 Instruct the server to send the whole response multiplexed, not just
332 the packfile section. All non-flush and non-delim PKT-LINE in the
333 response (not only in the packfile section) will then start with a byte
334 indicating its sideband (1, 2, or 3), and the server may send "0005\2"
335 (a PKT-LINE of sideband 2 with no payload) as a keepalive packet.
337 If the 'packfile-uris' feature is advertised, the following argument
338 can be included in the client's request as well as the potential
339 addition of the 'packfile-uris' section in the server's response as
342 packfile-uris <comma-separated list of protocols>
343 Indicates to the server that the client is willing to receive
344 URIs of any of the given protocols in place of objects in the
345 sent packfile. Before performing the connectivity check, the
346 client should download from all given URIs. Currently, the
347 protocols supported are "http" and "https".
349 The response of `fetch` is broken into a number of sections separated by
350 delimiter packets (0001), with each section beginning with its section
351 header. Most sections are sent only when the packfile is sent.
353 output = acknowledgements flush-pkt |
354 [acknowledgments delim-pkt] [shallow-info delim-pkt]
355 [wanted-refs delim-pkt] [packfile-uris delim-pkt]
358 acknowledgments = PKT-LINE("acknowledgments" LF)
361 ready = PKT-LINE("ready" LF)
362 nak = PKT-LINE("NAK" LF)
363 ack = PKT-LINE("ACK" SP obj-id LF)
365 shallow-info = PKT-LINE("shallow-info" LF)
366 *PKT-LINE((shallow | unshallow) LF)
367 shallow = "shallow" SP obj-id
368 unshallow = "unshallow" SP obj-id
370 wanted-refs = PKT-LINE("wanted-refs" LF)
371 *PKT-LINE(wanted-ref LF)
372 wanted-ref = obj-id SP refname
374 packfile-uris = PKT-LINE("packfile-uris" LF) *packfile-uri
375 packfile-uri = PKT-LINE(40*(HEXDIGIT) SP *%x20-ff LF)
377 packfile = PKT-LINE("packfile" LF)
378 *PKT-LINE(%x01-03 *%x00-ff)
380 acknowledgments section
381 * If the client determines that it is finished with negotiations by
382 sending a "done" line (thus requiring the server to send a packfile),
383 the acknowledgments sections MUST be omitted from the server's
386 * Always begins with the section header "acknowledgments"
388 * The server will respond with "NAK" if none of the object ids sent
389 as have lines were common.
391 * The server will respond with "ACK obj-id" for all of the
392 object ids sent as have lines which are common.
394 * A response cannot have both "ACK" lines as well as a "NAK"
397 * The server will respond with a "ready" line indicating that
398 the server has found an acceptable common base and is ready to
399 make and send a packfile (which will be found in the packfile
400 section of the same response)
402 * If the server has found a suitable cut point and has decided
403 to send a "ready" line, then the server can decide to (as an
404 optimization) omit any "ACK" lines it would have sent during
405 its response. This is because the server will have already
406 determined the objects it plans to send to the client and no
407 further negotiation is needed.
410 * If the client has requested a shallow fetch/clone, a shallow
411 client requests a fetch or the server is shallow then the
412 server's response may include a shallow-info section. The
413 shallow-info section will be included if (due to one of the
414 above conditions) the server needs to inform the client of any
415 shallow boundaries or adjustments to the clients already
416 existing shallow boundaries.
418 * Always begins with the section header "shallow-info"
420 * If a positive depth is requested, the server will compute the
421 set of commits which are no deeper than the desired depth.
423 * The server sends a "shallow obj-id" line for each commit whose
424 parents will not be sent in the following packfile.
426 * The server sends an "unshallow obj-id" line for each commit
427 which the client has indicated is shallow, but is no longer
428 shallow as a result of the fetch (due to its parents being
429 sent in the following packfile).
431 * The server MUST NOT send any "unshallow" lines for anything
432 which the client has not indicated was shallow as a part of
436 * This section is only included if the client has requested a
437 ref using a 'want-ref' line and if a packfile section is also
438 included in the response.
440 * Always begins with the section header "wanted-refs".
442 * The server will send a ref listing ("<oid> <refname>") for
443 each reference requested using 'want-ref' lines.
445 * The server MUST NOT send any refs which were not requested
446 using 'want-ref' lines.
448 packfile-uris section
449 * This section is only included if the client sent
450 'packfile-uris' and the server has at least one such URI to
453 * Always begins with the section header "packfile-uris".
455 * For each URI the server sends, it sends a hash of the pack's
456 contents (as output by git index-pack) followed by the URI.
458 * The hashes are 40 hex characters long. When Git upgrades to a new
459 hash algorithm, this might need to be updated. (It should match
460 whatever index-pack outputs after "pack\t" or "keep\t".
463 * This section is only included if the client has sent 'want'
464 lines in its request and either requested that no more
465 negotiation be done by sending 'done' or if the server has
466 decided it has found a sufficient cut point to produce a
469 * Always begins with the section header "packfile"
471 * The transmission of the packfile begins immediately after the
474 * The data transfer of the packfile is always multiplexed, using
475 the same semantics of the 'side-band-64k' capability from
476 protocol version 1. This means that each packet, during the
477 packfile data stream, is made up of a leading 4-byte pkt-line
478 length (typical of the pkt-line format), followed by a 1-byte
479 stream code, followed by the actual data.
481 The stream code can be one of:
483 2 - progress messages
484 3 - fatal error message just before stream aborts
489 If advertised, indicates that any number of server specific options can be
490 included in a request. This is done by sending each option as a
491 "server-option=<option>" capability line in the capability-list section of
494 The provided options must not contain a NUL or LF character.
499 The server can advertise the `object-format` capability with a value `X` (in the
500 form `object-format=X`) to notify the client that the server is able to deal
501 with objects using hash algorithm X. If not specified, the server is assumed to
502 only handle SHA-1. If the client would like to use a hash algorithm other than
503 SHA-1, it should specify its object-format string.
505 session-id=<session id>
506 ~~~~~~~~~~~~~~~~~~~~~~~
508 The server may advertise a session ID that can be used to identify this process
509 across multiple requests. The client may advertise its own session ID back to
512 Session IDs should be unique to a given process. They must fit within a
513 packet-line, and must not contain non-printable or whitespace characters. The
514 current implementation uses trace2 session IDs (see
515 link:api-trace2.html[api-trace2] for details), but this may change and users of
516 the session ID should not rely on this fact.