When "git fetch" encounters repositories with too many references, the
command line of "fetch-pack" that is run by a helper e.g. remote-curl, may
fail to hold all of them. Now such an internal invocation can feed the
references through the standard input of "fetch-pack".
By Ivan Todoroski
* it/fetch-pack-many-refs:
remote-curl: main test case for the OS command line overflow
fetch-pack: test cases for the new --stdin option
remote-curl: send the refs to fetch-pack on stdin
fetch-pack: new --stdin option to read refs from stdin
Conflicts:
t/t5500-fetch-pack.sh
If a remote repo has too many tags (or branches), cloning it over the
smart HTTP transport can fail because remote-curl.c puts all the refs
from the remote repo on the fetch-pack command line. This can make the
command line longer than the global OS command line limit, causing
fetch-pack to fail.
This is especially a problem on Windows where the command line limit is
orders of magnitude shorter than Linux. There are already real repos out
there that msysGit cannot clone over smart HTTP due to this problem.
Here is an easy way to trigger this problem:
git init too-many-refs
cd too-many-refs
echo bla > bla.txt
git add .
git commit -m test
sha=$(git rev-parse HEAD)
tag=$(perl -e 'print "bla" x 30')
for i in `seq 50000`; do
echo $sha refs/tags/$tag-$i >> .git/packed-refs
done
Then share this repo over the smart HTTP protocol and try cloning it:
$ git clone http://localhost/.../too-many-refs/.git
Cloning into 'too-many-refs'...
fatal: cannot exec 'fetch-pack': Argument list too long
50k tags is obviously an absurd number, but it is required to
demonstrate the problem on Linux because it has a much more generous
command line limit. On Windows the clone fails with as little as 500
tags in the above loop, which is getting uncomfortably close to the
number of tags you might see in real long lived repos.
This is not just theoretical, msysGit is already failing to clone our
company repo due to this. It's a large repo converted from CVS, nearly
10 years of history.
Four possible solutions were discussed on the Git mailing list (in no
particular order):
1) Call fetch-pack multiple times with smaller batches of refs.
This was dismissed as inefficient and inelegant.
2) Add option --refs-fd=$n to pass a an fd from where to read the refs.
This was rejected because inheriting descriptors other than
stdin/stdout/stderr through exec() is apparently problematic on Windows,
plus it would require changes to the run-command API to open extra
pipes.
3) Add option --refs-from=$tmpfile to pass the refs using a temp file.
This was not favored because of the temp file requirement.
4) Add option --stdin to pass the refs on stdin, one per line.
In the end this option was chosen as the most efficient and most
desirable from scripting perspective.
There was however a small complication when using stdin to pass refs to
fetch-pack. The --stateless-rpc option to fetch-pack also uses stdin for
communication with the remote server.
If we are going to sneak refs on stdin line by line, it would have to be
done very carefully in the presence of --stateless-rpc, because when
reading refs line by line we might read ahead too much data into our
buffer and eat some of the remote protocol data which is also coming on
stdin.
One way to solve this would be to refactor get_remote_heads() in
fetch-pack.c to accept a residual buffer from our stdin line parsing
above, but this function is used in several places so other callers
would be burdened by this residual buffer interface even when most of
them don't need it.
In the end we settled on the following solution:
If --stdin is specified without --stateless-rpc, fetch-pack would read
the refs from stdin one per line, in a script friendly format.
However if --stdin is specified together with --stateless-rpc,
fetch-pack would read the refs from stdin in packetized format
(pkt-line) with a flush packet terminating the list of refs. This way we
can read the exact number of bytes that we need from stdin, and then
get_remote_heads() can continue reading from the same fd without losing
a single byte of remote protocol data.
This way the --stdin option only loses generality and scriptability when
used together with --stateless-rpc, which is not easily scriptable
anyway because it also uses pkt-line when talking to the remote server.
Signed-off-by: Ivan Todoroski <grnch@gmx.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
By default, progress output is disabled if stderr is not a terminal.
The --progress option can be used to force progress output anyways.
Conversely, --no-progress does not force progress output. In particular,
if stderr is a terminal, progress output is enabled.
This is unintuitive. Change --no-progress to force output off.
Signed-off-by: Clemens Buchacher <drizzd@aon.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Objects in an alternate object database are already available to the
local repository and therefore don't need to be fetched. So mark them
as complete in everything_local().
This fixes a test in t5700.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The logic of the (single) caller is clearer without encapsulating this
one line in a function.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The parameters denote reference names, which are no longer 1:1 with
filesystem paths.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we are determining the list of refs to fetch via
fetch-pack, we have two sets of refs to compare: those on
the remote side, and a "match" list of things we want to
fetch. We iterate through the remote refs alphabetically,
seeing if each one is wanted by the "match" list.
Since def88e9 (Commit first cut at "git-fetch-pack",
2005-07-04), we have used the "path_match" function to do a
suffix match, where a remote ref is considered wanted if
any of the "match" elements is a suffix of the remote
refname.
This enables callers of fetch-pack to specify unqualified
refs and have them matched up with remote refs (e.g., ask
for "A" and get remote's "refs/heads/A"). However, if you
provide a fully qualified ref, then there are corner cases
where we provide the wrong answer. For example, given a
remote with two refs:
refs/foo/refs/heads/master
refs/heads/master
asking for "refs/heads/master" will first match
"refs/foo/refs/heads/master" by the suffix rule, and we will
erroneously fetch it instead of refs/heads/master.
As it turns out, all callers of fetch_pack do provide
fully-qualified refs for the match list. There are two ways
fetch_pack can get match lists:
1. Through the transport code (i.e., via git-fetch)
2. On the command-line of git-fetch-pack
In the first case, we will always be providing the names of
fully-qualified refs from "struct ref" objects. We will have
pre-matched those ref objects already (since we have to
handle more advanced matching, like wildcard refspecs), and
are just providing a list of the refs whose objects we need.
In the second case, users could in theory be providing
non-qualified refs on the command-line. However, the
fetch-pack documentation claims that refs should be fully
qualified (and has always done so since it was written in
2005).
Let's change this path_match call to simply check for string
equality, matching what the callers of fetch_pack are
expecting.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The get_remote_heads function reads the list of remote refs
during git protocol session. It dates all the way back to
def88e9 (Commit first cut at "git-fetch-pack", 2005-07-04).
At that time, the idea was to come up with a list of refs we
were interested in, and then filter the list as we got it
from the remote side.
Later, 1baaae5 (Make maximal use of the remote refs,
2005-10-28) stopped filtering at the get_remote_heads layer,
letting us use the non-matching refs to find common history.
As a result, all callers now simply pass an empty match
list (and any future callers will want to do the same). So
let's drop these now-useless parameters.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* mh/check-ref-format-3: (23 commits)
add_ref(): verify that the refname is formatted correctly
resolve_ref(): expand documentation
resolve_ref(): also treat a too-long SHA1 as invalid
resolve_ref(): emit warnings for improperly-formatted references
resolve_ref(): verify that the input refname has the right format
remote: avoid passing NULL to read_ref()
remote: use xstrdup() instead of strdup()
resolve_ref(): do not follow incorrectly-formatted symbolic refs
resolve_ref(): extract a function get_packed_ref()
resolve_ref(): turn buffer into a proper string as soon as possible
resolve_ref(): only follow a symlink that contains a valid, normalized refname
resolve_ref(): use prefixcmp()
resolve_ref(): explicitly fail if a symlink is not readable
Change check_refname_format() to reject unnormalized refnames
Inline function refname_format_print()
Make collapse_slashes() allocate memory for its result
Do not allow ".lock" at the end of any refname component
Refactor check_refname_format()
Change check_ref_format() to take a flags argument
Change bad_ref_char() to return a boolean value
...
Change check_ref_format() to take a flags argument that indicates what
is acceptable in the reference name (analogous to "git
check-ref-format"'s "--allow-onelevel" and "--refspec-pattern"). This
is more convenient for callers and also fixes a failure in the test
suite (and likely elsewhere in the code) by enabling "onelevel" and
"refspec-pattern" to be allowed independently of each other.
Also rename check_ref_format() to check_refname_format() to make it
obvious that it deals with refnames rather than references themselves.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This single variable can be used to set instead of setting fsckobjects
variable for fetch & receive independently.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This corresponds to receive.fsckobjects configuration variable added (a
lot) earlier in 20dc001 (receive-pack: allow using --strict mode for
unpacking objects, 2008-02-25).
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* nd/decorate-grafts:
log: Do not decorate replacements with --no-replace-objects
log: decorate "replaced" on to replaced commits
log: decorate grafted commits with "grafted"
Move write_shallow_commits to fetch-pack.c
Add for_each_commit_graft() to iterate all grafts
decoration: do not mis-decorate refs with same prefix
A malicious server can return ACK with non-existent SHA-1 or not a
commit. lookup_commit() in this case may return NULL. Do not let
fetch-pack crash by accessing NULL address in this case.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This function produces network traffic and should be in fetch-pack. It
has been in commit.c because it needs to iterate (private) graft
list. It can now do so using for_each_commit_graft().
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* jk/haves-from-alternate-odb:
receive-pack: eliminate duplicate .have refs
bisect: refactor sha1_array into a generic sha1 list
refactor refs_from_alternate_cb to allow passing extra data
The foreach_alt_odb function triggers a callback for each
alternate object db we have, with room for a single void
pointer as data. Currently, we always call refs_from_alternate_cb
as the callback function, and then pass another callback (to
receive each ref individually) as the void pointer.
This has two problems:
1. C technically forbids stuffing a function pointer into
a "void *". In practice, this probably doesn't matter
on any architectures git runs on, but it never hurts to
follow the letter of the law.
2. There is no room for an extra data pointer. Indeed, the
alternate_ref_fn that refs_from_alternate_cb calls
takes a void* for data, but we always pass it NULL.
Instead, let's properly stuff our function pointer into a
data struct, which also leaves room for an extra
caller-supplied data pointer. And to keep things simple for
existing callers, let's make a for_each_alternate_ref
function that takes care of creating the extra struct.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We add every local ref to a list so that we can mark them
and all of their ancestors back to a certain cutoff point.
However, if some refs point to the same commit, we will end
up adding them to the list many times.
Furthermore, since commit_lists are stored as linked lists,
we must do an O(n) traversal of the list in order to find
the right place to insert each commit. This makes building
the list O(n^2) in the number of refs.
For normal repositories, this isn't a big deal. We have a
few hundreds refs at most, and most of them are unique. But
consider an "alternates" repo that serves as an object
database for many other similar repos. For reachability, it
needs to keep a copy of the refs in each child repo. This
means it may have a large number of refs, many of which
point to the same commits.
By noting commits we have already added to the list, we can
shrink the size of "n" in such a repo to the number of
unique commits, which is on the order of what a normal repo
would contain (it's actually more than a normal repo, since child repos
may have branches at different states, but in practice it tends
to be much smaller than the list with duplicates).
Here are the results on one particular giant repo
(containing objects for all Rails forks on GitHub):
$ git for-each-ref | wc -l
112514
[before]
$ git fetch --no-tags ../remote.git
63.52user 0.12system 1:03.68elapsed 99%CPU (0avgtext+0avgdata 137648maxresident)k
1856inputs+48outputs (11major+19603minor)pagefaults 0swaps
$ git fetch --no-tags ../remote.git
6.15user 0.08system 0:06.25elapsed 99%CPU (0avgtext+0avgdata 123856maxresident)k
0inputs+40outputs (0major+18872minor)pagefaults 0swaps
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* sp/maint-fetch-pack-stop-early:
enable "no-done" extension only when fetching over smart-http
* sp/maint-upload-pack-stop-early:
enable "no-done" extension only when serving over smart-http
Last night I had to make these two emergency reverts, but now we have a
better understanding of which part of the topic was broken, let's get rid
of the revert to fix it correctly.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The fetch-pack/upload-pack protocol relies on the underlying transport
(local pipe or TCP socket) to have enough slack to allow one window worth
of data in flight without blocking the writer. Traditionally we always
relied on being able to have two windows of 32 "have"s in flight (roughly
3k bytes) to stream.
The recent "progressive-stride" change allows "fetch-pack" to send up to
1024 "have"s without reading any response from "upload-pack". The
outgoing pipe of "upload-pack" can be clogged with many ACK and NAK that
are unread, while "fetch-pack" is still stuffing its outgoing pipe with
more "have"s, leading to a deadlock.
Revert the change unless we are in stateless rpc (aka smart-http) mode, as
using a large window full of "have"s is still a good way to help reduce
the number of back-and-forth, and there is no buffering issue there (it is
strictly "ping-pong" without an overlap).
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When 'no-done' protocol extension is used, the upload-pack (i.e. the
server side) process stops listening to the fetch-pack after issuing the
final NAK, and starts sending the generated pack data back, but there may
be more "have" send by the latter in flight that the fetch-pack is
expecting to be responded with ACK/NAK. This will typically result in a
deadlock (both will block on write that the other end never reads) or
SIGPIPE on the fetch-pack end (upload-pack will finish writing a small
pack and goes away).
Disable it unless fetch-pack is running under smart-http, where there is
no such streaming issue.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
* jc/fetch-progressive-stride:
fetch-pack: use smaller handshake window for initial request
fetch-pack: progressively use larger handshake windows
fetch-pack: factor out hardcoded handshake window size
Conflicts:
builtin/fetch-pack.c
* jc/maint-fetch-alt:
fetch-pack: objects in our alternates are available to us
refs_from_alternate: helper to use refs from alternates
Conflicts:
builtin/receive-pack.c
Fix warnings from 'make check'.
- These files don't include 'builtin.h' causing sparse to complain that
cmd_* isn't declared:
builtin/clone.c:364, builtin/fetch-pack.c:797,
builtin/fmt-merge-msg.c:34, builtin/hash-object.c:78,
builtin/merge-index.c:69, builtin/merge-recursive.c:22
builtin/merge-tree.c:341, builtin/mktag.c:156, builtin/notes.c:426
builtin/notes.c:822, builtin/pack-redundant.c:596,
builtin/pack-refs.c:10, builtin/patch-id.c:60, builtin/patch-id.c:149,
builtin/remote.c:1512, builtin/remote-ext.c:240,
builtin/remote-fd.c:53, builtin/reset.c:236, builtin/send-pack.c:384,
builtin/unpack-file.c:25, builtin/var.c:75
- These files have symbols which should be marked static since they're
only file scope:
submodule.c:12, diff.c:631, replace_object.c:92, submodule.c:13,
submodule.c:14, trace.c:78, transport.c:195, transport-helper.c:79,
unpack-trees.c:19, url.c:3, url.c:18, url.c:104, url.c:117, url.c:123,
url.c:129, url.c:136, thread-utils.c:21, thread-utils.c:48
- These files redeclare symbols to be different types:
builtin/index-pack.c:210, parse-options.c:564, parse-options.c:571,
usage.c:49, usage.c:58, usage.c:63, usage.c:72
- These files use a literal integer 0 when they really should use a NULL
pointer:
daemon.c:663, fast-import.c:2942, imap-send.c:1072, notes-merge.c:362
While we're in the area, clean up some unused #includes in builtin files
(mostly exec_cmd.h).
Signed-off-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Start the initial request small by halving the INITIAL_FLUSH (we will try
to stay one window ahead of the server, so we would end up giving twice as
many "have" in flight at the very beginning). We may want to tweak these
values even more, taking MTU into account.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn Pearce <spearce@spearce.org>
The client has to dig the history deeper when more recent parts of its
history do not have any overlap with the server it is fetching from. Make
the handshake window exponentially larger as we dig deeper, with a
reasonable upper cap.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn Pearce <spearce@spearce.org>
The "git fetch" client presents the most recent 32 commits it has to the
server and gives a chance to the server to say "ok, we heard enough", and
continues reporting what it has in chunks of 32 commits, digging its
history down to older commits.
Move the hardcoded size of the handshake window outside the code, so that
we can tweak it more easily.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn Pearce <spearce@spearce.org>
Use the helper function split from the receiving end of "git push" to
allow the same optimization on the receiving end of "git fetch".
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
If enabled on the connection "multi_ack_detailed no-done" as a
pair allows the remote upload-pack process to send a PACK down
to the client as soon as a "ACK %s ready" message was also sent.
Over git:// and ssh:// where a bi-directional stream is in place
this has very little difference over the classical version that
waits for the client to send a "done\n" line by itself. It does
slightly reduce the latency involved to start the pack stream as
there is one less round-trip from client->server required.
Over smart HTTP this avoids needing to send a final RPC that has
all of the prior common objects. Instead the server is able to
return a pack as soon as its ready to. For many common users the
smart HTTP fetch is now just 2 requests: GET .../info/refs, and
a POST .../git-upload-pack to not only negotiate but also receive
the pack stream. Only users who have more than 32 local unshared
commits with the remote will need additional requests to negotiate
a common merge base.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If multi_ack_detailed was selected in the protocol capabilities
(both client and server are >= Git 1.6.6) the upload-pack side will
send "ACK %s ready" when it knows how to safely cut the graph and
produce a reasonable pack for the want list that was already sent
on the connection.
Upon receiving "ACK %s ready" there is no point in looking at
the remaining commits inside of rev_list. Sending additional
"have %s" lines to the remote will not construct a smaller pack.
It is unlikely a commit older than the current cut point will have
a better delta base than the cut point itself has.
The original design of this code had fetch-pack empty rev_list by
marking a commit and its transitive ancestors COMMON whenever the
remote side said "ACK %s {continue,common}" and skipping over any
already COMMON commits during get_rev(). This approach does not
work when most of rev_list is actually COMMON_REF, commits that
are pointed to by a reference on the remote, which exist locally,
and which have not yet been sent to the remote as a "have %s" line.
Most of the common references are tags in the ref/tags namespace,
using points in the commit graph that are more than 1 commit apart.
In git.git itself, this is currently 340 tags, 339 of which point to
commits in the commit graph. fetch-pack pushes all of these into
rev_list, but is unable to mark them COMMON and discard during a
remote's "ACK %s {continue,common}" because it does not parse through
the entire parent chain. Not parsing the entire parent chain is
an optimization to avoid walking back to the roots of the repository.
Assuming the client is only following the remote (and does not make
its own local commits), the client needs 11 rounds to spin through
the entire list of tags (32 commits per round, ceil(339/32) == 11).
Unfortunately the server knows on the first "have %s" line that
it can produce a good pack, and does not need to see the remaining
320 tags in the other 10 rounds.
Over git:// and ssh:// this isn't as bad as it sounds, the client is
only transmitting an extra 16,000 bytes that it doesn't need to send.
Over smart HTTP, the client must do an additional 10 HTTP POST
requests, each of which incurs round-trip latency, and must upload
the entire state vector of all known common objects. On the final
POST request, this is 16 KiB worth of data.
Fix all of this by clearing rev_list as soon as the remote side
says it can construct a pack.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This shows a trace of all packets coming in or out of a given
program. This can help with debugging object negotiation or
other protocol issues.
To keep the code changes simple, we operate at the lowest
level, meaning we don't necessarily understand what's in the
packets. The one exception is a packet starting with "PACK",
which causes us to skip that packet and turn off tracing
(since the gigantic pack data will not be interesting to
read, at least not in the trace format).
We show both written and read packets. In the local case,
this may mean you will see packets twice (written by the
sender and read by the receiver). However, for cases where
the other end is remote, this allows you to see the full
conversation.
Packet tracing can be enabled with GIT_TRACE_PACKET=<foo>,
where <foo> takes the same arguments as GIT_TRACE.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add commit_list prefix to insert_by_date function and to sort_by_date,
so it's clear that these functions refer to commit_list structure.
Signed-off-by: Thiago Farina <tfransosi@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This shrinks the top-level directory a bit, and makes it much more
pleasant to use auto-completion on the thing. Instead of
[torvalds@nehalem git]$ em buil<tab>
Display all 180 possibilities? (y or n)
[torvalds@nehalem git]$ em builtin-sh
builtin-shortlog.c builtin-show-branch.c builtin-show-ref.c
builtin-shortlog.o builtin-show-branch.o builtin-show-ref.o
[torvalds@nehalem git]$ em builtin-shor<tab>
builtin-shortlog.c builtin-shortlog.o
[torvalds@nehalem git]$ em builtin-shortlog.c
you get
[torvalds@nehalem git]$ em buil<tab> [type]
builtin/ builtin.h
[torvalds@nehalem git]$ em builtin [auto-completes to]
[torvalds@nehalem git]$ em builtin/sh<tab> [type]
shortlog.c shortlog.o show-branch.c show-branch.o show-ref.c show-ref.o
[torvalds@nehalem git]$ em builtin/sho [auto-completes to]
[torvalds@nehalem git]$ em builtin/shor<tab> [type]
shortlog.c shortlog.o
[torvalds@nehalem git]$ em builtin/shortlog.c
which doesn't seem all that different, but not having that annoying
break in "Display all 180 possibilities?" is quite a relief.
NOTE! If you do this in a clean tree (no object files etc), or using an
editor that has auto-completion rules that ignores '*.o' files, you
won't see that annoying 'Display all 180 possibilities?' message - it
will just show the choices instead. I think bash has some cut-off
around 100 choices or something.
So the reason I see this is that I'm using an odd editory, and thus
don't have the rules to cut down on auto-completion. But you can
simulate that by using 'ls' instead, or something similar.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
