Merge branch 'js/hashmap-update-sample'
[git/git.git] / Documentation / technical / pack-protocol.txt
1 Packfile transfer protocols
2 ===========================
4 Git supports transferring data in packfiles over the ssh://, git://, http:// and
5 file:// transports. There exist two sets of protocols, one for pushing
6 data from a client to a server and another for fetching data from a
7 server to a client. The three transports (ssh, git, file) use the same
8 protocol to transfer data. http is documented in http-protocol.txt.
10 The processes invoked in the canonical Git implementation are 'upload-pack'
11 on the server side and 'fetch-pack' on the client side for fetching data;
12 then 'receive-pack' on the server and 'send-pack' on the client for pushing
13 data. The protocol functions to have a server tell a client what is
14 currently on the server, then for the two to negotiate the smallest amount
15 of data to send in order to fully update one or the other.
17 pkt-line Format
18 ---------------
20 The descriptions below build on the pkt-line format described in
21 protocol-common.txt. When the grammar indicate `PKT-LINE(...)`, unless
22 otherwise noted the usual pkt-line LF rules apply: the sender SHOULD
23 include a LF, but the receiver MUST NOT complain if it is not present.
25 Transports
26 ----------
27 There are three transports over which the packfile protocol is
28 initiated. The Git transport is a simple, unauthenticated server that
29 takes the command (almost always 'upload-pack', though Git
30 servers can be configured to be globally writable, in which 'receive-
31 pack' initiation is also allowed) with which the client wishes to
32 communicate and executes it and connects it to the requesting
33 process.
35 In the SSH transport, the client just runs the 'upload-pack'
36 or 'receive-pack' process on the server over the SSH protocol and then
37 communicates with that invoked process over the SSH connection.
39 The file:// transport runs the 'upload-pack' or 'receive-pack'
40 process locally and communicates with it over a pipe.
42 Extra Parameters
43 ----------------
45 The protocol provides a mechanism in which clients can send additional
46 information in its first message to the server. These are called "Extra
47 Parameters", and are supported by the Git, SSH, and HTTP protocols.
49 Each Extra Parameter takes the form of `<key>=<value>` or `<key>`.
51 Servers that receive any such Extra Parameters MUST ignore all
52 unrecognized keys. Currently, the only Extra Parameter recognized is
53 "version=1".
55 Git Transport
56 -------------
58 The Git transport starts off by sending the command and repository
59 on the wire using the pkt-line format, followed by a NUL byte and a
60 hostname parameter, terminated by a NUL byte.
62 0033git-upload-pack /project.git\\0
64 The transport may send Extra Parameters by adding an additional NUL
65 byte, and then adding one or more NUL-terminated strings:
67 003egit-upload-pack /project.git\\0\0version=1\0
69 --
70 git-proto-request = request-command SP pathname NUL
71 [ host-parameter NUL ] [ NUL extra-parameters ]
72 request-command = "git-upload-pack" / "git-receive-pack" /
73 "git-upload-archive" ; case sensitive
74 pathname = *( %x01-ff ) ; exclude NUL
75 host-parameter = "host=" hostname [ ":" port ]
76 extra-parameters = 1*extra-parameter
77 extra-parameter = 1*( %x01-ff ) NUL
78 --
80 host-parameter is used for the
81 git-daemon name based virtual hosting. See --interpolated-path
82 option to git daemon, with the %H/%CH format characters.
84 Basically what the Git client is doing to connect to an 'upload-pack'
85 process on the server side over the Git protocol is this:
87 $ echo -e -n \
88 "0039git-upload-pack /schacon/gitbook.git\\0" |
89 nc -v 9418
91 If the server refuses the request for some reasons, it could abort
92 gracefully with an error message.
94 ----
95 error-line = PKT-LINE("ERR" SP explanation-text)
96 ----
99 SSH Transport
100 -------------
102 Initiating the upload-pack or receive-pack processes over SSH is
103 executing the binary on the server via SSH remote execution.
104 It is basically equivalent to running this:
106 $ ssh "git-upload-pack '/project.git'"
108 For a server to support Git pushing and pulling for a given user over
109 SSH, that user needs to be able to execute one or both of those
110 commands via the SSH shell that they are provided on login. On some
111 systems, that shell access is limited to only being able to run those
112 two commands, or even just one of them.
114 In an ssh:// format URI, it's absolute in the URI, so the '/' after
115 the host name (or port number) is sent as an argument, which is then
116 read by the remote git-upload-pack exactly as is, so it's effectively
117 an absolute path in the remote filesystem.
119 git clone ssh://
120 |
121 v
122 ssh "git-upload-pack '/project.git'"
124 In a "user@host:path" format URI, its relative to the user's home
125 directory, because the Git client will run:
127 git clone
128 |
129 v
130 ssh "git-upload-pack 'project.git'"
132 The exception is if a '~' is used, in which case
133 we execute it without the leading '/'.
135 ssh://,
136 |
137 v
138 ssh "git-upload-pack '~alice/project.git'"
140 Depending on the value of the `protocol.version` configuration variable,
141 Git may attempt to send Extra Parameters as a colon-separated string in
142 the GIT_PROTOCOL environment variable. This is done only if
143 the `ssh.variant` configuration variable indicates that the ssh command
144 supports passing environment variables as an argument.
146 A few things to remember here:
148 - The "command name" is spelled with dash (e.g. git-upload-pack), but
149 this can be overridden by the client;
151 - The repository path is always quoted with single quotes.
153 Fetching Data From a Server
154 ---------------------------
156 When one Git repository wants to get data that a second repository
157 has, the first can 'fetch' from the second. This operation determines
158 what data the server has that the client does not then streams that
159 data down to the client in packfile format.
162 Reference Discovery
163 -------------------
165 When the client initially connects the server will immediately respond
166 with a version number (if "version=1" is sent as an Extra Parameter),
167 and a listing of each reference it has (all branches and tags) along
168 with the object name that each reference currently points to.
170 $ echo -e -n "0044git-upload-pack /schacon/gitbook.git\\0\0version=1\0" |
171 nc -v 9418
172 000aversion 1
173 00887217a7c7e582c46cec22a130adf4b9d7d950fba0 HEAD\0multi_ack thin-pack
174 side-band side-band-64k ofs-delta shallow no-progress include-tag
175 00441d3fcd5ced445d1abc402225c0b8a1299641f497 refs/heads/integration
176 003f7217a7c7e582c46cec22a130adf4b9d7d950fba0 refs/heads/master
177 003cb88d2441cac0977faf98efc80305012112238d9d refs/tags/v0.9
178 003c525128480b96c89e6418b1e40909bf6c5b2d580f refs/tags/v1.0
179 003fe92df48743b7bc7d26bcaabfddde0a1e20cae47c refs/tags/v1.0^{}
180 0000
182 The returned response is a pkt-line stream describing each ref and
183 its current value. The stream MUST be sorted by name according to
184 the C locale ordering.
186 If HEAD is a valid ref, HEAD MUST appear as the first advertised
187 ref. If HEAD is not a valid ref, HEAD MUST NOT appear in the
188 advertisement list at all, but other refs may still appear.
190 The stream MUST include capability declarations behind a NUL on the
191 first ref. The peeled value of a ref (that is "ref^{}") MUST be
192 immediately after the ref itself, if presented. A conforming server
193 MUST peel the ref if it's an annotated tag.
195 ----
196 advertised-refs = *1("version 1")
197 (no-refs / list-of-refs)
198 *shallow
199 flush-pkt
201 no-refs = PKT-LINE(zero-id SP "capabilities^{}"
202 NUL capability-list)
204 list-of-refs = first-ref *other-ref
205 first-ref = PKT-LINE(obj-id SP refname
206 NUL capability-list)
208 other-ref = PKT-LINE(other-tip / other-peeled)
209 other-tip = obj-id SP refname
210 other-peeled = obj-id SP refname "^{}"
212 shallow = PKT-LINE("shallow" SP obj-id)
214 capability-list = capability *(SP capability)
215 capability = 1*(LC_ALPHA / DIGIT / "-" / "_")
216 LC_ALPHA = %x61-7A
217 ----
219 Server and client MUST use lowercase for obj-id, both MUST treat obj-id
220 as case-insensitive.
222 See protocol-capabilities.txt for a list of allowed server capabilities
223 and descriptions.
225 Packfile Negotiation
226 --------------------
227 After reference and capabilities discovery, the client can decide to
228 terminate the connection by sending a flush-pkt, telling the server it can
229 now gracefully terminate, and disconnect, when it does not need any pack
230 data. This can happen with the ls-remote command, and also can happen when
231 the client already is up to date.
233 Otherwise, it enters the negotiation phase, where the client and
234 server determine what the minimal packfile necessary for transport is,
235 by telling the server what objects it wants, its shallow objects
236 (if any), and the maximum commit depth it wants (if any). The client
237 will also send a list of the capabilities it wants to be in effect,
238 out of what the server said it could do with the first 'want' line.
240 ----
241 upload-request = want-list
242 *shallow-line
243 *1depth-request
244 flush-pkt
246 want-list = first-want
247 *additional-want
249 shallow-line = PKT-LINE("shallow" SP obj-id)
251 depth-request = PKT-LINE("deepen" SP depth) /
252 PKT-LINE("deepen-since" SP timestamp) /
253 PKT-LINE("deepen-not" SP ref)
255 first-want = PKT-LINE("want" SP obj-id SP capability-list)
256 additional-want = PKT-LINE("want" SP obj-id)
258 depth = 1*DIGIT
259 ----
261 Clients MUST send all the obj-ids it wants from the reference
262 discovery phase as 'want' lines. Clients MUST send at least one
263 'want' command in the request body. Clients MUST NOT mention an
264 obj-id in a 'want' command which did not appear in the response
265 obtained through ref discovery.
267 The client MUST write all obj-ids which it only has shallow copies
268 of (meaning that it does not have the parents of a commit) as
269 'shallow' lines so that the server is aware of the limitations of
270 the client's history.
272 The client now sends the maximum commit history depth it wants for
273 this transaction, which is the number of commits it wants from the
274 tip of the history, if any, as a 'deepen' line. A depth of 0 is the
275 same as not making a depth request. The client does not want to receive
276 any commits beyond this depth, nor does it want objects needed only to
277 complete those commits. Commits whose parents are not received as a
278 result are defined as shallow and marked as such in the server. This
279 information is sent back to the client in the next step.
281 Once all the 'want's and 'shallow's (and optional 'deepen') are
282 transferred, clients MUST send a flush-pkt, to tell the server side
283 that it is done sending the list.
285 Otherwise, if the client sent a positive depth request, the server
286 will determine which commits will and will not be shallow and
287 send this information to the client. If the client did not request
288 a positive depth, this step is skipped.
290 ----
291 shallow-update = *shallow-line
292 *unshallow-line
293 flush-pkt
295 shallow-line = PKT-LINE("shallow" SP obj-id)
297 unshallow-line = PKT-LINE("unshallow" SP obj-id)
298 ----
300 If the client has requested a positive depth, the server will compute
301 the set of commits which are no deeper than the desired depth. The set
302 of commits start at the client's wants.
304 The server writes 'shallow' lines for each
305 commit whose parents will not be sent as a result. The server writes
306 an 'unshallow' line for each commit which the client has indicated is
307 shallow, but is no longer shallow at the currently requested depth
308 (that is, its parents will now be sent). The server MUST NOT mark
309 as unshallow anything which the client has not indicated was shallow.
311 Now the client will send a list of the obj-ids it has using 'have'
312 lines, so the server can make a packfile that only contains the objects
313 that the client needs. In multi_ack mode, the canonical implementation
314 will send up to 32 of these at a time, then will send a flush-pkt. The
315 canonical implementation will skip ahead and send the next 32 immediately,
316 so that there is always a block of 32 "in-flight on the wire" at a time.
318 ----
319 upload-haves = have-list
320 compute-end
322 have-list = *have-line
323 have-line = PKT-LINE("have" SP obj-id)
324 compute-end = flush-pkt / PKT-LINE("done")
325 ----
327 If the server reads 'have' lines, it then will respond by ACKing any
328 of the obj-ids the client said it had that the server also has. The
329 server will ACK obj-ids differently depending on which ack mode is
330 chosen by the client.
332 In multi_ack mode:
334 * the server will respond with 'ACK obj-id continue' for any common
335 commits.
337 * once the server has found an acceptable common base commit and is
338 ready to make a packfile, it will blindly ACK all 'have' obj-ids
339 back to the client.
341 * the server will then send a 'NAK' and then wait for another response
342 from the client - either a 'done' or another list of 'have' lines.
344 In multi_ack_detailed mode:
346 * the server will differentiate the ACKs where it is signaling
347 that it is ready to send data with 'ACK obj-id ready' lines, and
348 signals the identified common commits with 'ACK obj-id common' lines.
350 Without either multi_ack or multi_ack_detailed:
352 * upload-pack sends "ACK obj-id" on the first common object it finds.
353 After that it says nothing until the client gives it a "done".
355 * upload-pack sends "NAK" on a flush-pkt if no common object
356 has been found yet. If one has been found, and thus an ACK
357 was already sent, it's silent on the flush-pkt.
359 After the client has gotten enough ACK responses that it can determine
360 that the server has enough information to send an efficient packfile
361 (in the canonical implementation, this is determined when it has received
362 enough ACKs that it can color everything left in the --date-order queue
363 as common with the server, or the --date-order queue is empty), or the
364 client determines that it wants to give up (in the canonical implementation,
365 this is determined when the client sends 256 'have' lines without getting
366 any of them ACKed by the server - meaning there is nothing in common and
367 the server should just send all of its objects), then the client will send
368 a 'done' command. The 'done' command signals to the server that the client
369 is ready to receive its packfile data.
371 However, the 256 limit *only* turns on in the canonical client
372 implementation if we have received at least one "ACK %s continue"
373 during a prior round. This helps to ensure that at least one common
374 ancestor is found before we give up entirely.
376 Once the 'done' line is read from the client, the server will either
377 send a final 'ACK obj-id' or it will send a 'NAK'. 'obj-id' is the object
378 name of the last commit determined to be common. The server only sends
379 ACK after 'done' if there is at least one common base and multi_ack or
380 multi_ack_detailed is enabled. The server always sends NAK after 'done'
381 if there is no common base found.
383 Instead of 'ACK' or 'NAK', the server may send an error message (for
384 example, if it does not recognize an object in a 'want' line received
385 from the client).
387 Then the server will start sending its packfile data.
389 ----
390 server-response = *ack_multi ack / nak / error-line
391 ack_multi = PKT-LINE("ACK" SP obj-id ack_status)
392 ack_status = "continue" / "common" / "ready"
393 ack = PKT-LINE("ACK" SP obj-id)
394 nak = PKT-LINE("NAK")
395 error-line = PKT-LINE("ERR" SP explanation-text)
396 ----
398 A simple clone may look like this (with no 'have' lines):
400 ----
401 C: 0054want 74730d410fcb6603ace96f1dc55ea6196122532d multi_ack \
402 side-band-64k ofs-delta\n
403 C: 0032want 7d1665144a3a975c05f1f43902ddaf084e784dbe\n
404 C: 0032want 5a3f6be755bbb7deae50065988cbfa1ffa9ab68a\n
405 C: 0032want 7e47fe2bd8d01d481f44d7af0531bd93d3b21c01\n
406 C: 0032want 74730d410fcb6603ace96f1dc55ea6196122532d\n
407 C: 0000
408 C: 0009done\n
410 S: 0008NAK\n
412 ----
414 An incremental update (fetch) response might look like this:
416 ----
417 C: 0054want 74730d410fcb6603ace96f1dc55ea6196122532d multi_ack \
418 side-band-64k ofs-delta\n
419 C: 0032want 7d1665144a3a975c05f1f43902ddaf084e784dbe\n
420 C: 0032want 5a3f6be755bbb7deae50065988cbfa1ffa9ab68a\n
421 C: 0000
422 C: 0032have 7e47fe2bd8d01d481f44d7af0531bd93d3b21c01\n
423 C: [30 more have lines]
424 C: 0032have 74730d410fcb6603ace96f1dc55ea6196122532d\n
425 C: 0000
427 S: 003aACK 7e47fe2bd8d01d481f44d7af0531bd93d3b21c01 continue\n
428 S: 003aACK 74730d410fcb6603ace96f1dc55ea6196122532d continue\n
429 S: 0008NAK\n
431 C: 0009done\n
433 S: 0031ACK 74730d410fcb6603ace96f1dc55ea6196122532d\n
435 ----
438 Packfile Data
439 -------------
441 Now that the client and server have finished negotiation about what
442 the minimal amount of data that needs to be sent to the client is, the server
443 will construct and send the required data in packfile format.
445 See pack-format.txt for what the packfile itself actually looks like.
447 If 'side-band' or 'side-band-64k' capabilities have been specified by
448 the client, the server will send the packfile data multiplexed.
450 Each packet starting with the packet-line length of the amount of data
451 that follows, followed by a single byte specifying the sideband the
452 following data is coming in on.
454 In 'side-band' mode, it will send up to 999 data bytes plus 1 control
455 code, for a total of up to 1000 bytes in a pkt-line. In 'side-band-64k'
456 mode it will send up to 65519 data bytes plus 1 control code, for a
457 total of up to 65520 bytes in a pkt-line.
459 The sideband byte will be a '1', '2' or a '3'. Sideband '1' will contain
460 packfile data, sideband '2' will be used for progress information that the
461 client will generally print to stderr and sideband '3' is used for error
462 information.
464 If no 'side-band' capability was specified, the server will stream the
465 entire packfile without multiplexing.
468 Pushing Data To a Server
469 ------------------------
471 Pushing data to a server will invoke the 'receive-pack' process on the
472 server, which will allow the client to tell it which references it should
473 update and then send all the data the server will need for those new
474 references to be complete. Once all the data is received and validated,
475 the server will then update its references to what the client specified.
477 Authentication
478 --------------
480 The protocol itself contains no authentication mechanisms. That is to be
481 handled by the transport, such as SSH, before the 'receive-pack' process is
482 invoked. If 'receive-pack' is configured over the Git transport, those
483 repositories will be writable by anyone who can access that port (9418) as
484 that transport is unauthenticated.
486 Reference Discovery
487 -------------------
489 The reference discovery phase is done nearly the same way as it is in the
490 fetching protocol. Each reference obj-id and name on the server is sent
491 in packet-line format to the client, followed by a flush-pkt. The only
492 real difference is that the capability listing is different - the only
493 possible values are 'report-status', 'delete-refs', 'ofs-delta' and
494 'push-options'.
496 Reference Update Request and Packfile Transfer
497 ----------------------------------------------
499 Once the client knows what references the server is at, it can send a
500 list of reference update requests. For each reference on the server
501 that it wants to update, it sends a line listing the obj-id currently on
502 the server, the obj-id the client would like to update it to and the name
503 of the reference.
505 This list is followed by a flush-pkt.
507 ----
508 update-requests = *shallow ( command-list | push-cert )
510 shallow = PKT-LINE("shallow" SP obj-id)
512 command-list = PKT-LINE(command NUL capability-list)
513 *PKT-LINE(command)
514 flush-pkt
516 command = create / delete / update
517 create = zero-id SP new-id SP name
518 delete = old-id SP zero-id SP name
519 update = old-id SP new-id SP name
521 old-id = obj-id
522 new-id = obj-id
524 push-cert = PKT-LINE("push-cert" NUL capability-list LF)
525 PKT-LINE("certificate version 0.1" LF)
526 PKT-LINE("pusher" SP ident LF)
527 PKT-LINE("pushee" SP url LF)
528 PKT-LINE("nonce" SP nonce LF)
529 *PKT-LINE("push-option" SP push-option LF)
531 *PKT-LINE(command LF)
532 *PKT-LINE(gpg-signature-lines LF)
533 PKT-LINE("push-cert-end" LF)
535 push-option = 1*( VCHAR | SP )
536 ----
538 If the server has advertised the 'push-options' capability and the client has
539 specified 'push-options' as part of the capability list above, the client then
540 sends its push options followed by a flush-pkt.
542 ----
543 push-options = *PKT-LINE(push-option) flush-pkt
544 ----
546 For backwards compatibility with older Git servers, if the client sends a push
547 cert and push options, it MUST send its push options both embedded within the
548 push cert and after the push cert. (Note that the push options within the cert
549 are prefixed, but the push options after the cert are not.) Both these lists
550 MUST be the same, modulo the prefix.
552 After that the packfile that
553 should contain all the objects that the server will need to complete the new
554 references will be sent.
556 ----
557 packfile = "PACK" 28*(OCTET)
558 ----
560 If the receiving end does not support delete-refs, the sending end MUST
561 NOT ask for delete command.
563 If the receiving end does not support push-cert, the sending end
564 MUST NOT send a push-cert command. When a push-cert command is
565 sent, command-list MUST NOT be sent; the commands recorded in the
566 push certificate is used instead.
568 The packfile MUST NOT be sent if the only command used is 'delete'.
570 A packfile MUST be sent if either create or update command is used,
571 even if the server already has all the necessary objects. In this
572 case the client MUST send an empty packfile. The only time this
573 is likely to happen is if the client is creating
574 a new branch or a tag that points to an existing obj-id.
576 The server will receive the packfile, unpack it, then validate each
577 reference that is being updated that it hasn't changed while the request
578 was being processed (the obj-id is still the same as the old-id), and
579 it will run any update hooks to make sure that the update is acceptable.
580 If all of that is fine, the server will then update the references.
582 Push Certificate
583 ----------------
585 A push certificate begins with a set of header lines. After the
586 header and an empty line, the protocol commands follow, one per
587 line. Note that the trailing LF in push-cert PKT-LINEs is _not_
588 optional; it must be present.
590 Currently, the following header fields are defined:
592 `pusher` ident::
593 Identify the GPG key in "Human Readable Name <email@address>"
594 format.
596 `pushee` url::
597 The repository URL (anonymized, if the URL contains
598 authentication material) the user who ran `git push`
599 intended to push into.
601 `nonce` nonce::
602 The 'nonce' string the receiving repository asked the
603 pushing user to include in the certificate, to prevent
604 replay attacks.
606 The GPG signature lines are a detached signature for the contents
607 recorded in the push certificate before the signature block begins.
608 The detached signature is used to certify that the commands were
609 given by the pusher, who must be the signer.
611 Report Status
612 -------------
614 After receiving the pack data from the sender, the receiver sends a
615 report if 'report-status' capability is in effect.
616 It is a short listing of what happened in that update. It will first
617 list the status of the packfile unpacking as either 'unpack ok' or
618 'unpack [error]'. Then it will list the status for each of the references
619 that it tried to update. Each line is either 'ok [refname]' if the
620 update was successful, or 'ng [refname] [error]' if the update was not.
622 ----
623 report-status = unpack-status
624 1*(command-status)
625 flush-pkt
627 unpack-status = PKT-LINE("unpack" SP unpack-result)
628 unpack-result = "ok" / error-msg
630 command-status = command-ok / command-fail
631 command-ok = PKT-LINE("ok" SP refname)
632 command-fail = PKT-LINE("ng" SP refname SP error-msg)
634 error-msg = 1*(OCTECT) ; where not "ok"
635 ----
637 Updates can be unsuccessful for a number of reasons. The reference can have
638 changed since the reference discovery phase was originally sent, meaning
639 someone pushed in the meantime. The reference being pushed could be a
640 non-fast-forward reference and the update hooks or configuration could be
641 set to not allow that, etc. Also, some references can be updated while others
642 can be rejected.
644 An example client/server communication might look like this:
646 ----
647 S: 007c74730d410fcb6603ace96f1dc55ea6196122532d refs/heads/local\0report-status delete-refs ofs-delta\n
648 S: 003e7d1665144a3a975c05f1f43902ddaf084e784dbe refs/heads/debug\n
649 S: 003f74730d410fcb6603ace96f1dc55ea6196122532d refs/heads/master\n
650 S: 003f74730d410fcb6603ace96f1dc55ea6196122532d refs/heads/team\n
651 S: 0000
653 C: 003e7d1665144a3a975c05f1f43902ddaf084e784dbe 74730d410fcb6603ace96f1dc55ea6196122532d refs/heads/debug\n
654 C: 003e74730d410fcb6603ace96f1dc55ea6196122532d 5a3f6be755bbb7deae50065988cbfa1ffa9ab68a refs/heads/master\n
655 C: 0000
658 S: 000eunpack ok\n
659 S: 0018ok refs/heads/debug\n
660 S: 002ang refs/heads/master non-fast-forward\n
661 ----