[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 Git Transport
43 -------------
45 The Git transport starts off by sending the command and repository
46 on the wire using the pkt-line format, followed by a NUL byte and a
47 hostname parameter, terminated by a NUL byte.
49 0032git-upload-pack /project.git\0host=myserver.com\0
51 --
52 git-proto-request = request-command SP pathname NUL [ host-parameter NUL ]
53 request-command = "git-upload-pack" / "git-receive-pack" /
54 "git-upload-archive" ; case sensitive
55 pathname = *( %x01-ff ) ; exclude NUL
56 host-parameter = "host=" hostname [ ":" port ]
57 --
59 Only host-parameter is allowed in the git-proto-request. Clients
60 MUST NOT attempt to send additional parameters. It is used for the
61 git-daemon name based virtual hosting. See --interpolated-path
62 option to git daemon, with the %H/%CH format characters.
64 Basically what the Git client is doing to connect to an 'upload-pack'
65 process on the server side over the Git protocol is this:
67 $ echo -e -n \
68 "0039git-upload-pack /schacon/gitbook.git\0host=example.com\0" |
69 nc -v example.com 9418
71 If the server refuses the request for some reasons, it could abort
72 gracefully with an error message.
74 ----
75 error-line = PKT-LINE("ERR" SP explanation-text)
76 ----
79 SSH Transport
80 -------------
82 Initiating the upload-pack or receive-pack processes over SSH is
83 executing the binary on the server via SSH remote execution.
84 It is basically equivalent to running this:
86 $ ssh git.example.com "git-upload-pack '/project.git'"
88 For a server to support Git pushing and pulling for a given user over
89 SSH, that user needs to be able to execute one or both of those
90 commands via the SSH shell that they are provided on login. On some
91 systems, that shell access is limited to only being able to run those
92 two commands, or even just one of them.
94 In an ssh:// format URI, it's absolute in the URI, so the '/' after
95 the host name (or port number) is sent as an argument, which is then
96 read by the remote git-upload-pack exactly as is, so it's effectively
97 an absolute path in the remote filesystem.
99 git clone ssh://user@example.com/project.git
100 |
101 v
102 ssh user@example.com "git-upload-pack '/project.git'"
104 In a "user@host:path" format URI, its relative to the user's home
105 directory, because the Git client will run:
107 git clone user@example.com:project.git
108 |
109 v
110 ssh user@example.com "git-upload-pack 'project.git'"
112 The exception is if a '~' is used, in which case
113 we execute it without the leading '/'.
115 ssh://user@example.com/~alice/project.git,
116 |
117 v
118 ssh user@example.com "git-upload-pack '~alice/project.git'"
120 A few things to remember here:
122 - The "command name" is spelled with dash (e.g. git-upload-pack), but
123 this can be overridden by the client;
125 - The repository path is always quoted with single quotes.
127 Fetching Data From a Server
128 ---------------------------
130 When one Git repository wants to get data that a second repository
131 has, the first can 'fetch' from the second. This operation determines
132 what data the server has that the client does not then streams that
133 data down to the client in packfile format.
136 Reference Discovery
137 -------------------
139 When the client initially connects the server will immediately respond
140 with a listing of each reference it has (all branches and tags) along
141 with the object name that each reference currently points to.
143 $ echo -e -n "0039git-upload-pack /schacon/gitbook.git\0host=example.com\0" |
144 nc -v example.com 9418
145 00887217a7c7e582c46cec22a130adf4b9d7d950fba0 HEAD\0multi_ack thin-pack
146 side-band side-band-64k ofs-delta shallow no-progress include-tag
147 00441d3fcd5ced445d1abc402225c0b8a1299641f497 refs/heads/integration
148 003f7217a7c7e582c46cec22a130adf4b9d7d950fba0 refs/heads/master
149 003cb88d2441cac0977faf98efc80305012112238d9d refs/tags/v0.9
150 003c525128480b96c89e6418b1e40909bf6c5b2d580f refs/tags/v1.0
151 003fe92df48743b7bc7d26bcaabfddde0a1e20cae47c refs/tags/v1.0^{}
152 0000
154 The returned response is a pkt-line stream describing each ref and
155 its current value. The stream MUST be sorted by name according to
156 the C locale ordering.
158 If HEAD is a valid ref, HEAD MUST appear as the first advertised
159 ref. If HEAD is not a valid ref, HEAD MUST NOT appear in the
160 advertisement list at all, but other refs may still appear.
162 The stream MUST include capability declarations behind a NUL on the
163 first ref. The peeled value of a ref (that is "ref^{}") MUST be
164 immediately after the ref itself, if presented. A conforming server
165 MUST peel the ref if it's an annotated tag.
167 ----
168 advertised-refs = (no-refs / list-of-refs)
169 *shallow
170 flush-pkt
172 no-refs = PKT-LINE(zero-id SP "capabilities^{}"
173 NUL capability-list)
175 list-of-refs = first-ref *other-ref
176 first-ref = PKT-LINE(obj-id SP refname
177 NUL capability-list)
179 other-ref = PKT-LINE(other-tip / other-peeled)
180 other-tip = obj-id SP refname
181 other-peeled = obj-id SP refname "^{}"
183 shallow = PKT-LINE("shallow" SP obj-id)
185 capability-list = capability *(SP capability)
186 capability = 1*(LC_ALPHA / DIGIT / "-" / "_")
187 LC_ALPHA = %x61-7A
188 ----
190 Server and client MUST use lowercase for obj-id, both MUST treat obj-id
191 as case-insensitive.
193 See protocol-capabilities.txt for a list of allowed server capabilities
194 and descriptions.
196 Packfile Negotiation
197 --------------------
198 After reference and capabilities discovery, the client can decide to
199 terminate the connection by sending a flush-pkt, telling the server it can
200 now gracefully terminate, and disconnect, when it does not need any pack
201 data. This can happen with the ls-remote command, and also can happen when
202 the client already is up-to-date.
204 Otherwise, it enters the negotiation phase, where the client and
205 server determine what the minimal packfile necessary for transport is,
206 by telling the server what objects it wants, its shallow objects
207 (if any), and the maximum commit depth it wants (if any). The client
208 will also send a list of the capabilities it wants to be in effect,
209 out of what the server said it could do with the first 'want' line.
211 ----
212 upload-request = want-list
213 *shallow-line
214 *1depth-request
215 flush-pkt
217 want-list = first-want
218 *additional-want
220 shallow-line = PKT-LINE("shallow" SP obj-id)
222 depth-request = PKT-LINE("deepen" SP depth)
224 first-want = PKT-LINE("want" SP obj-id SP capability-list)
225 additional-want = PKT-LINE("want" SP obj-id)
227 depth = 1*DIGIT
228 ----
230 Clients MUST send all the obj-ids it wants from the reference
231 discovery phase as 'want' lines. Clients MUST send at least one
232 'want' command in the request body. Clients MUST NOT mention an
233 obj-id in a 'want' command which did not appear in the response
234 obtained through ref discovery.
236 The client MUST write all obj-ids which it only has shallow copies
237 of (meaning that it does not have the parents of a commit) as
238 'shallow' lines so that the server is aware of the limitations of
239 the client's history.
241 The client now sends the maximum commit history depth it wants for
242 this transaction, which is the number of commits it wants from the
243 tip of the history, if any, as a 'deepen' line. A depth of 0 is the
244 same as not making a depth request. The client does not want to receive
245 any commits beyond this depth, nor does it want objects needed only to
246 complete those commits. Commits whose parents are not received as a
247 result are defined as shallow and marked as such in the server. This
248 information is sent back to the client in the next step.
250 Once all the 'want's and 'shallow's (and optional 'deepen') are
251 transferred, clients MUST send a flush-pkt, to tell the server side
252 that it is done sending the list.
254 Otherwise, if the client sent a positive depth request, the server
255 will determine which commits will and will not be shallow and
256 send this information to the client. If the client did not request
257 a positive depth, this step is skipped.
259 ----
260 shallow-update = *shallow-line
261 *unshallow-line
262 flush-pkt
264 shallow-line = PKT-LINE("shallow" SP obj-id)
266 unshallow-line = PKT-LINE("unshallow" SP obj-id)
267 ----
269 If the client has requested a positive depth, the server will compute
270 the set of commits which are no deeper than the desired depth. The set
271 of commits start at the client's wants.
273 The server writes 'shallow' lines for each
274 commit whose parents will not be sent as a result. The server writes
275 an 'unshallow' line for each commit which the client has indicated is
276 shallow, but is no longer shallow at the currently requested depth
277 (that is, its parents will now be sent). The server MUST NOT mark
278 as unshallow anything which the client has not indicated was shallow.
280 Now the client will send a list of the obj-ids it has using 'have'
281 lines, so the server can make a packfile that only contains the objects
282 that the client needs. In multi_ack mode, the canonical implementation
283 will send up to 32 of these at a time, then will send a flush-pkt. The
284 canonical implementation will skip ahead and send the next 32 immediately,
285 so that there is always a block of 32 "in-flight on the wire" at a time.
287 ----
288 upload-haves = have-list
289 compute-end
291 have-list = *have-line
292 have-line = PKT-LINE("have" SP obj-id)
293 compute-end = flush-pkt / PKT-LINE("done")
294 ----
296 If the server reads 'have' lines, it then will respond by ACKing any
297 of the obj-ids the client said it had that the server also has. The
298 server will ACK obj-ids differently depending on which ack mode is
299 chosen by the client.
301 In multi_ack mode:
303 * the server will respond with 'ACK obj-id continue' for any common
304 commits.
306 * once the server has found an acceptable common base commit and is
307 ready to make a packfile, it will blindly ACK all 'have' obj-ids
308 back to the client.
310 * the server will then send a 'NACK' and then wait for another response
311 from the client - either a 'done' or another list of 'have' lines.
313 In multi_ack_detailed mode:
315 * the server will differentiate the ACKs where it is signaling
316 that it is ready to send data with 'ACK obj-id ready' lines, and
317 signals the identified common commits with 'ACK obj-id common' lines.
319 Without either multi_ack or multi_ack_detailed:
321 * upload-pack sends "ACK obj-id" on the first common object it finds.
322 After that it says nothing until the client gives it a "done".
324 * upload-pack sends "NAK" on a flush-pkt if no common object
325 has been found yet. If one has been found, and thus an ACK
326 was already sent, it's silent on the flush-pkt.
328 After the client has gotten enough ACK responses that it can determine
329 that the server has enough information to send an efficient packfile
330 (in the canonical implementation, this is determined when it has received
331 enough ACKs that it can color everything left in the --date-order queue
332 as common with the server, or the --date-order queue is empty), or the
333 client determines that it wants to give up (in the canonical implementation,
334 this is determined when the client sends 256 'have' lines without getting
335 any of them ACKed by the server - meaning there is nothing in common and
336 the server should just send all of its objects), then the client will send
337 a 'done' command. The 'done' command signals to the server that the client
338 is ready to receive its packfile data.
340 However, the 256 limit *only* turns on in the canonical client
341 implementation if we have received at least one "ACK %s continue"
342 during a prior round. This helps to ensure that at least one common
343 ancestor is found before we give up entirely.
345 Once the 'done' line is read from the client, the server will either
346 send a final 'ACK obj-id' or it will send a 'NAK'. 'obj-id' is the object
347 name of the last commit determined to be common. The server only sends
348 ACK after 'done' if there is at least one common base and multi_ack or
349 multi_ack_detailed is enabled. The server always sends NAK after 'done'
350 if there is no common base found.
352 Then the server will start sending its packfile data.
354 ----
355 server-response = *ack_multi ack / nak
356 ack_multi = PKT-LINE("ACK" SP obj-id ack_status)
357 ack_status = "continue" / "common" / "ready"
358 ack = PKT-LINE("ACK" SP obj-id)
359 nak = PKT-LINE("NAK")
360 ----
362 A simple clone may look like this (with no 'have' lines):
364 ----
365 C: 0054want 74730d410fcb6603ace96f1dc55ea6196122532d multi_ack \
366 side-band-64k ofs-delta\n
367 C: 0032want 7d1665144a3a975c05f1f43902ddaf084e784dbe\n
368 C: 0032want 5a3f6be755bbb7deae50065988cbfa1ffa9ab68a\n
369 C: 0032want 7e47fe2bd8d01d481f44d7af0531bd93d3b21c01\n
370 C: 0032want 74730d410fcb6603ace96f1dc55ea6196122532d\n
371 C: 0000
372 C: 0009done\n
374 S: 0008NAK\n
376 ----
378 An incremental update (fetch) response might look like this:
380 ----
381 C: 0054want 74730d410fcb6603ace96f1dc55ea6196122532d multi_ack \
382 side-band-64k ofs-delta\n
383 C: 0032want 7d1665144a3a975c05f1f43902ddaf084e784dbe\n
384 C: 0032want 5a3f6be755bbb7deae50065988cbfa1ffa9ab68a\n
385 C: 0000
386 C: 0032have 7e47fe2bd8d01d481f44d7af0531bd93d3b21c01\n
387 C: [30 more have lines]
388 C: 0032have 74730d410fcb6603ace96f1dc55ea6196122532d\n
389 C: 0000
391 S: 003aACK 7e47fe2bd8d01d481f44d7af0531bd93d3b21c01 continue\n
392 S: 003aACK 74730d410fcb6603ace96f1dc55ea6196122532d continue\n
393 S: 0008NAK\n
395 C: 0009done\n
397 S: 0031ACK 74730d410fcb6603ace96f1dc55ea6196122532d\n
399 ----
402 Packfile Data
403 -------------
405 Now that the client and server have finished negotiation about what
406 the minimal amount of data that needs to be sent to the client is, the server
407 will construct and send the required data in packfile format.
409 See pack-format.txt for what the packfile itself actually looks like.
411 If 'side-band' or 'side-band-64k' capabilities have been specified by
412 the client, the server will send the packfile data multiplexed.
414 Each packet starting with the packet-line length of the amount of data
415 that follows, followed by a single byte specifying the sideband the
416 following data is coming in on.
418 In 'side-band' mode, it will send up to 999 data bytes plus 1 control
419 code, for a total of up to 1000 bytes in a pkt-line. In 'side-band-64k'
420 mode it will send up to 65519 data bytes plus 1 control code, for a
421 total of up to 65520 bytes in a pkt-line.
423 The sideband byte will be a '1', '2' or a '3'. Sideband '1' will contain
424 packfile data, sideband '2' will be used for progress information that the
425 client will generally print to stderr and sideband '3' is used for error
426 information.
428 If no 'side-band' capability was specified, the server will stream the
429 entire packfile without multiplexing.
432 Pushing Data To a Server
433 ------------------------
435 Pushing data to a server will invoke the 'receive-pack' process on the
436 server, which will allow the client to tell it which references it should
437 update and then send all the data the server will need for those new
438 references to be complete. Once all the data is received and validated,
439 the server will then update its references to what the client specified.
441 Authentication
442 --------------
444 The protocol itself contains no authentication mechanisms. That is to be
445 handled by the transport, such as SSH, before the 'receive-pack' process is
446 invoked. If 'receive-pack' is configured over the Git transport, those
447 repositories will be writable by anyone who can access that port (9418) as
448 that transport is unauthenticated.
450 Reference Discovery
451 -------------------
453 The reference discovery phase is done nearly the same way as it is in the
454 fetching protocol. Each reference obj-id and name on the server is sent
455 in packet-line format to the client, followed by a flush-pkt. The only
456 real difference is that the capability listing is different - the only
457 possible values are 'report-status', 'delete-refs' and 'ofs-delta'.
459 Reference Update Request and Packfile Transfer
460 ----------------------------------------------
462 Once the client knows what references the server is at, it can send a
463 list of reference update requests. For each reference on the server
464 that it wants to update, it sends a line listing the obj-id currently on
465 the server, the obj-id the client would like to update it to and the name
466 of the reference.
468 This list is followed by a flush-pkt and then the packfile that should
469 contain all the objects that the server will need to complete the new
470 references.
472 ----
473 update-request = *shallow ( command-list | push-cert ) [packfile]
475 shallow = PKT-LINE("shallow" SP obj-id)
477 command-list = PKT-LINE(command NUL capability-list)
478 *PKT-LINE(command)
479 flush-pkt
481 command = create / delete / update
482 create = zero-id SP new-id SP name
483 delete = old-id SP zero-id SP name
484 update = old-id SP new-id SP name
486 old-id = obj-id
487 new-id = obj-id
489 push-cert = PKT-LINE("push-cert" NUL capability-list LF)
490 PKT-LINE("certificate version 0.1" LF)
491 PKT-LINE("pusher" SP ident LF)
492 PKT-LINE("pushee" SP url LF)
493 PKT-LINE("nonce" SP nonce LF)
495 *PKT-LINE(command LF)
496 *PKT-LINE(gpg-signature-lines LF)
497 PKT-LINE("push-cert-end" LF)
499 packfile = "PACK" 28*(OCTET)
500 ----
502 If the receiving end does not support delete-refs, the sending end MUST
503 NOT ask for delete command.
505 If the receiving end does not support push-cert, the sending end
506 MUST NOT send a push-cert command. When a push-cert command is
507 sent, command-list MUST NOT be sent; the commands recorded in the
508 push certificate is used instead.
510 The packfile MUST NOT be sent if the only command used is 'delete'.
512 A packfile MUST be sent if either create or update command is used,
513 even if the server already has all the necessary objects. In this
514 case the client MUST send an empty packfile. The only time this
515 is likely to happen is if the client is creating
516 a new branch or a tag that points to an existing obj-id.
518 The server will receive the packfile, unpack it, then validate each
519 reference that is being updated that it hasn't changed while the request
520 was being processed (the obj-id is still the same as the old-id), and
521 it will run any update hooks to make sure that the update is acceptable.
522 If all of that is fine, the server will then update the references.
524 Push Certificate
525 ----------------
527 A push certificate begins with a set of header lines. After the
528 header and an empty line, the protocol commands follow, one per
529 line. Note that the the trailing LF in push-cert PKT-LINEs is _not_
530 optional; it must be present.
532 Currently, the following header fields are defined:
534 `pusher` ident::
535 Identify the GPG key in "Human Readable Name <email@address>"
536 format.
538 `pushee` url::
539 The repository URL (anonymized, if the URL contains
540 authentication material) the user who ran `git push`
541 intended to push into.
543 `nonce` nonce::
544 The 'nonce' string the receiving repository asked the
545 pushing user to include in the certificate, to prevent
546 replay attacks.
548 The GPG signature lines are a detached signature for the contents
549 recorded in the push certificate before the signature block begins.
550 The detached signature is used to certify that the commands were
551 given by the pusher, who must be the signer.
553 Report Status
554 -------------
556 After receiving the pack data from the sender, the receiver sends a
557 report if 'report-status' capability is in effect.
558 It is a short listing of what happened in that update. It will first
559 list the status of the packfile unpacking as either 'unpack ok' or
560 'unpack [error]'. Then it will list the status for each of the references
561 that it tried to update. Each line is either 'ok [refname]' if the
562 update was successful, or 'ng [refname] [error]' if the update was not.
564 ----
565 report-status = unpack-status
566 1*(command-status)
567 flush-pkt
569 unpack-status = PKT-LINE("unpack" SP unpack-result)
570 unpack-result = "ok" / error-msg
572 command-status = command-ok / command-fail
573 command-ok = PKT-LINE("ok" SP refname)
574 command-fail = PKT-LINE("ng" SP refname SP error-msg)
576 error-msg = 1*(OCTECT) ; where not "ok"
577 ----
579 Updates can be unsuccessful for a number of reasons. The reference can have
580 changed since the reference discovery phase was originally sent, meaning
581 someone pushed in the meantime. The reference being pushed could be a
582 non-fast-forward reference and the update hooks or configuration could be
583 set to not allow that, etc. Also, some references can be updated while others
584 can be rejected.
586 An example client/server communication might look like this:
588 ----
589 S: 007c74730d410fcb6603ace96f1dc55ea6196122532d refs/heads/local\0report-status delete-refs ofs-delta\n
590 S: 003e7d1665144a3a975c05f1f43902ddaf084e784dbe refs/heads/debug\n
591 S: 003f74730d410fcb6603ace96f1dc55ea6196122532d refs/heads/master\n
592 S: 003f74730d410fcb6603ace96f1dc55ea6196122532d refs/heads/team\n
593 S: 0000
595 C: 003e7d1665144a3a975c05f1f43902ddaf084e784dbe 74730d410fcb6603ace96f1dc55ea6196122532d refs/heads/debug\n
596 C: 003e74730d410fcb6603ace96f1dc55ea6196122532d 5a3f6be755bbb7deae50065988cbfa1ffa9ab68a refs/heads/master\n
597 C: 0000
600 S: 000eunpack ok\n
601 S: 0018ok refs/heads/debug\n
602 S: 002ang refs/heads/master non-fast-forward\n
603 ----