Serving a "git fetch" client over "git://" and "ssh://" protocols
using the on-wire protocol version 2 was buggy on the server end
when the client needs to make a follow-up request to
e.g. auto-follow tags.
* cc/upload-pack-v2-fetch-fix:
upload-pack: clear filter_options for each v2 fetch command
Because of the request/response model of protocol v2, the
upload_pack_v2() function is sometimes called twice in the same
process, while 'struct list_objects_filter_options filter_options'
was declared as static at the beginning of 'upload-pack.c'.
This made the check in list_objects_filter_die_if_populated(), which
is called by process_args(), fail the second time upload_pack_v2() is
called, as filter_options had already been populated the first time.
To fix that, filter_options is not static any more. It's now owned
directly by upload_pack(). It's now also part of 'struct
upload_pack_data', so that it's owned indirectly by upload_pack_v2().
In the long term, the goal is to also have upload_pack() use
'struct upload_pack_data', so adding filter_options to this struct
makes more sense than to have it owned directly by upload_pack_v2().
This fixes the first of the 2 bugs documented by d0badf8797
(partial-clone: demonstrate bugs in partial fetch, 2020-02-21).
Helped-by: Derrick Stolee <dstolee@microsoft.com>
Helped-by: Jeff King <peff@peff.net>
Helped-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are many functions in commit.h that are more related to shallow
repositories than they are to any sort of generic commit machinery.
Likely this began when there were only a few shallow-related functions,
and commit.h seemed a reasonable enough place to put them.
But, now there are a good number of shallow-related functions, and
placing them all in 'commit.h' doesn't make sense.
This patch extracts a 'shallow.h', which takes all of the declarations
from 'commit.h' for functions which already exist in 'shallow.c'. We
will bring the remaining shallow-related functions defined in 'commit.c'
in a subsequent patch.
For now, move only the ones that already are implemented in 'shallow.c',
and update the necessary includes.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When processing the arguments list for a v2 ls-refs or fetch command, we
loop like this:
while (packet_reader_read(request) != PACKET_READ_FLUSH) {
const char *arg = request->line;
...handle arg...
}
to read and handle packets until we see a flush. The hidden assumption
here is that anything except PACKET_READ_FLUSH will give us valid packet
data to read. But that's not true; PACKET_READ_DELIM or PACKET_READ_EOF
will leave packet->line as NULL, and we'll segfault trying to look at
it.
Instead, we should follow the more careful model demonstrated on the
client side (e.g., in process_capabilities_v2): keep looping as long
as we get normal packets, and then make sure that we broke out of the
loop due to a real flush. That fixes the segfault and correctly
diagnoses any unexpected input from the client.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Previously when iterating through git config variables, worktree config
and local config were both considered "CONFIG_SCOPE_REPO". This was
never a problem before as no one had needed to differentiate between the
two cases, but future functionality may care whether or not the config
options come from a worktree or from the repository's actual local
config file. For example, the planned feature to add a '--show-scope'
to config to allow a user to see which scope listed config options come
from would confuse users if it just printed 'repo' rather than 'local'
or 'worktree' as the documentation would lead them to expect. As well
as the additional benefit of making the implementation look more like
how the documentation describes the interface.
To accomplish this we split out what was previously considered repo
scope to be local and worktree.
The clients of 'current_config_scope()' who cared about
CONFIG_SCOPE_REPO are also modified to similarly care about
CONFIG_SCOPE_WORKTREE and CONFIG_SCOPE_LOCAL to preserve previous behavior.
Signed-off-by: Matthew Rogers <mattr94@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "upload-pack" (the counterpart of "git fetch") needs to disable
commit-graph when responding to a shallow clone/fetch request, but
the way this was done made Git panic, which has been corrected.
* jk/disable-commit-graph-during-upload-pack:
upload-pack: disable commit graph more gently for shallow traversal
commit-graph: bump DIE_ON_LOAD check to actual load-time
The list-objects-filter API (used to create a sparse/lazy clone)
learned to take a combined filter specification.
* md/list-objects-filter-combo:
list-objects-filter-options: make parser void
list-objects-filter-options: clean up use of ALLOC_GROW
list-objects-filter-options: allow mult. --filter
strbuf: give URL-encoding API a char predicate fn
list-objects-filter-options: make filter_spec a string_list
list-objects-filter-options: move error check up
list-objects-filter: implement composite filters
list-objects-filter-options: always supply *errbuf
list-objects-filter: put omits set in filter struct
list-objects-filter: encapsulate filter components
When the client has asked for certain shallow options like
"deepen-since", we do a custom rev-list walk that pretends to be
shallow. Before doing so, we have to disable the commit-graph, since it
is not compatible with the shallow view of the repository. That's
handled by 829a321569 (commit-graph: close_commit_graph before shallow
walk, 2018-08-20). That commit literally closes and frees our
repo->objects->commit_graph struct.
That creates an interesting problem for commits that have _already_ been
parsed using the commit graph. Their commit->object.parsed flag is set,
their commit->graph_pos is set, but their commit->maybe_tree may still
be NULL. When somebody later calls repo_get_commit_tree(), we see that
we haven't loaded the tree oid yet and try to get it from the commit
graph. But since it has been freed, we segfault!
So the root of the issue is a data dependency between the commit's
lazy-load of the tree oid and the fact that the commit graph can go
away mid-process. How can we resolve it?
There are a couple of general approaches:
1. The obvious answer is to avoid loading the tree from the graph when
we see that it's NULL. But then what do we return for the tree oid?
If we return NULL, our caller in do_traverse() will rightly
complain that we have no tree. We'd have to fallback to loading the
actual commit object and re-parsing it. That requires teaching
parse_commit_buffer() to understand re-parsing (i.e., not starting
from a clean slate and not leaking any allocated bits like parent
list pointers).
2. When we close the commit graph, walk through the set of in-memory
objects and clear any graph_pos pointers. But this means we also
have to "unparse" any such commits so that we know they still need
to open the commit object to fill in their trees. So it's no less
complicated than (1), and is more expensive (since we clear objects
we might not later need).
3. Stop freeing the commit-graph struct. Continue to let it be used
for lazy-loads of tree oids, but let upload-pack specify that it
shouldn't be used for further commit parsing.
4. Push the whole shallow rev-list out to its own sub-process, with
the commit-graph disabled from the start, giving it a clean memory
space to work from.
I've chosen (3) here. Options (1) and (2) would work, but are
non-trivial to implement. Option (4) is more expensive, and I'm not sure
how complicated it is (shelling out for the actual rev-list part is
easy, but we do then parse the resulting commits internally, and I'm not
clear which parts need to be handling shallow-ness).
The new test in t5500 triggers this segfault, but see the comments there
for how horribly intimate it has to be with how both upload-pack and
commit graphs work.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Code clean-up to remove hardcoded SHA-1 hash from many places.
* jk/oidhash:
hashmap: convert sha1hash() to oidhash()
hash.h: move object_id definition from cache.h
khash: rename oid helper functions
khash: drop sha1-specific map types
pack-bitmap: convert khash_sha1 maps into kh_oid_map
delta-islands: convert island_marks khash to use oids
khash: rename kh_oid_t to kh_oid_set
khash: drop broken oid_map typedef
object: convert create_object() to use object_id
object: convert internal hash_obj() to object_id
object: convert lookup_object() to use object_id
object: convert lookup_unknown_object() to use object_id
pack-objects: convert locate_object_entry_hash() to object_id
pack-objects: convert packlist_find() to use object_id
pack-bitmap-write: convert some helpers to use object_id
upload-pack: rename a "sha1" variable to "oid"
describe: fix accidental oid/hash type-punning
The commit-graph file is now part of the "files that the runtime
may keep open file descriptors on, all of which would need to be
closed when done with the object store", and the file descriptor to
an existing commit-graph file now is closed before "gc" finalizes a
new instance to replace it.
* ds/close-object-store:
packfile: rename close_all_packs to close_object_store
packfile: close commit-graph in close_all_packs
commit-graph: use raw_object_store when closing
Allow combining of multiple filters by simply repeating the --filter
flag. Before this patch, the user had to combine them in a single flag
somewhat awkwardly (e.g. --filter=combine:FOO+BAR), including
URL-encoding the individual filters.
To make this work, in the --filter flag parsing callback, rather than
error out when we detect that the filter_options struct is already
populated, we modify it in-place to contain the added sub-filter. The
existing sub-filter becomes the lhs of the combined filter, and the
next sub-filter becomes the rhs. We also have to URL-encode the LHS and
RHS sub-filters.
We can simplify the operation if the LHS is already a combine: filter.
In that case, we just append the URL-encoded RHS sub-filter to the LHS
spec to get the new spec.
Helped-by: Emily Shaffer <emilyshaffer@google.com>
Helped-by: Jeff Hostetler <git@jeffhostetler.com>
Helped-by: Jeff King <peff@peff.net>
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Matthew DeVore <matvore@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Make the filter_spec string a string_list rather than a raw C string.
The list of strings must be concatted together to make a complete
filter_spec. A future patch will use this capability to build "combine:"
filter specs gradually.
A strbuf would seem to be a more natural choice for this object, but it
unfortunately requires initialization besides just zero'ing out the
memory. This results in all container structs, and all containers of
those structs, etc., to also require initialization. Initializing them
all would be more cumbersome that simply using a string_list, which
behaves properly when its contents are zero'd.
For the purposes of code simplification, change behavior in how filter
specs are conveyed over the protocol: do not normalize the tree:<depth>
filter specs since there should be no server in existence that supports
tree:# but not tree:#k etc.
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Matthew DeVore <matvore@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are no callers left of lookup_object() that aren't just passing us
the "hash" member of a "struct object_id". Let's take the whole struct,
which gets us closer to removing all raw sha1 variables. It also
matches the existing conversions of lookup_blob(), etc.
The conversions of callers were done by hand, but they're all mechanical
one-liners.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are no callers left of lookup_unknown_object() that aren't just
passing us the "hash" member of a "struct object_id". Let's take the
whole struct, which gets us closer to removing all raw sha1 variables.
It also matches the existing conversions of lookup_blob(), etc.
The conversions of callers were done by hand, but they're all mechanical
one-liners.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This variable is a "struct object_id", but uses the old-style name
"sha1". Let's call it oid to match more modern code (and make it clear
that it can handle any algorithm, since it uses parse_oid_hex()
properly).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The server side support for "git fetch" used to show incorrect
value for the HEAD symbolic ref when the namespace feature is in
use, which has been corrected.
* jk/HEAD-symref-in-xfer-namespaces:
upload-pack: strip namespace from symref data
The close_commit_graph() method took a repository struct, but then
only uses the raw_object_store within. Change the function prototype
to make the method more flexible.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 7171d8c15f (upload-pack: send symbolic ref information as
capability, 2013-09-17), we've sent cloning and fetching clients special
information about which branch HEAD is pointing to, so that they don't
have to guess based on matching up commit ids.
However, this feature has never worked properly with the GIT_NAMESPACE
feature. Because upload-pack uses head_ref_namespaced(find_symref), we
do find and report on refs/namespaces/foo/HEAD instead of the actual
HEAD of the repo. This makes sense, since the branch pointed to by the
top-level HEAD may not be advertised at all. But we do two things wrong:
1. We report the full name refs/namespaces/foo/HEAD, instead of just
HEAD. Meaning no client is going to bother doing anything with that
symref, since we're not otherwise advertising it.
2. We report the symref destination using its full name (e.g.,
refs/namespaces/foo/refs/heads/master). That's similarly useless to
the client, who only saw "refs/heads/master" in the advertisement.
We should be stripping the namespace prefix off of both places (which
this patch fixes).
Likely nobody noticed because we tend to do the right thing anyway. Bug
(1) means that we said nothing about HEAD (just refs/namespace/foo/HEAD).
And so the client half of the code, from a45b5f0552 (connect: annotate
refs with their symref information in get_remote_head(), 2013-09-17),
does not annotate HEAD, and we use the fallback in guess_remote_head(),
matching refs by object id. Which is usually right. It only falls down
in ambiguous cases, like the one laid out in the included test.
This also means that we don't have to worry about breaking anybody who
was putting pre-stripped names into their namespace symrefs when we fix
bug (2). Because of bug (1), nobody would have been using the symref we
advertised in the first place (not to mention that those symrefs would
have appeared broken for any non-namespaced access).
Note that we have separate fixes here for the v0 and v2 protocols. The
symref advertisement moved in v2 to be a part of the ls-refs command.
This actually gets part (1) right, since the symref annotation
piggy-backs on the existing ref advertisement, which is properly
stripped. But it still needs a fix for part (2). The included tests
cover both protocols.
Reported-by: Bryan Turner <bturner@atlassian.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
On a filesystem like HFS+, the names of the refs stored as filesystem
entities may become different from what the end-user expects, just
like files in the working tree get "renamed". Work around the
mismatch by paying attention to the core.precomposeUnicode
configuration.
* en/unicode-in-refnames:
Honor core.precomposeUnicode in more places
Further code clean-up to allow the lowest level of name-to-object
mapping layer to work with a passed-in repository other than the
default one.
* nd/sha1-name-c-wo-the-repository: (34 commits)
sha1-name.c: remove the_repo from get_oid_mb()
sha1-name.c: remove the_repo from other get_oid_*
sha1-name.c: remove the_repo from maybe_die_on_misspelt_object_name
submodule-config.c: use repo_get_oid for reading .gitmodules
sha1-name.c: add repo_get_oid()
sha1-name.c: remove the_repo from get_oid_with_context_1()
sha1-name.c: remove the_repo from resolve_relative_path()
sha1-name.c: remove the_repo from diagnose_invalid_index_path()
sha1-name.c: remove the_repo from handle_one_ref()
sha1-name.c: remove the_repo from get_oid_1()
sha1-name.c: remove the_repo from get_oid_basic()
sha1-name.c: remove the_repo from get_describe_name()
sha1-name.c: remove the_repo from get_oid_oneline()
sha1-name.c: add repo_interpret_branch_name()
sha1-name.c: remove the_repo from interpret_branch_mark()
sha1-name.c: remove the_repo from interpret_nth_prior_checkout()
sha1-name.c: remove the_repo from get_short_oid()
sha1-name.c: add repo_for_each_abbrev()
sha1-name.c: store and use repo in struct disambiguate_state
sha1-name.c: add repo_find_unique_abbrev_r()
...
On Mac's HFS where git sets core.precomposeUnicode to true automatically
by git init/clone, when a user creates a simple unicode refname (in NFC
format) such as españa:
$ git branch españa
different commands would display the branch name differently. For
example, git branch, git log --decorate, and git fast-export all used
65 73 70 61 c3 b1 61 (or "espa\xc3\xb1a")
(NFC form) while show-ref would use
65 73 70 61 6e cc 83 61 (or "espan\xcc\x83a")
(NFD form). A stress test for git filter-repo was tripped up by this
inconsistency, though digging in I found that the problems could
compound; for example, if the user ran
$ git pack-refs --all
and then tried to check out the branch, they would be met with:
$ git checkout españa
error: pathspec 'españa' did not match any file(s) known to git
$ git checkout españa --
fatal: invalid reference: españa
$ git branch
españa
* master
Note that the user could run the `git branch` command first and copy and
paste the `españa` portion of the output and still see the same two
errors. Also, if the user added --no-prune to the pack-refs command,
then they would see three branches: master, españa, and españa (those
last two are NFC vs. NFD forms, even if they render the same).
Further, if the user had the `españa` branch checked out before
running `git pack-refs --all`, the user would be greeted with (note
that I'm trimming trailing output with an ellipsis):
$ git rev-parse HEAD
fatal: ambiguous argument 'HEAD': unknown revision or path...
$ git status
On branch españa
No commits yet...
Or worse, if the user didn't check this stuff first, running `git
commit` will create a new commit with all changes of all of history
being squashed into it.
In addition to pack-refs, one could also get into this state with
upload-pack or anything that calls either pack-refs or upload-pack (e.g.
gc or clone).
Add code in a few places (pack-refs, show-ref, upload-pack) to check and
honor the setting of core.precomposeUnicode to avoid these bugs.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Commit bdb31eada7 (upload-pack: report "not our ref" to client,
2017-02-23) catches the case where a client asks for an object we don't
have, and issues a message that the client can show to the user (in
addition to dying and writing to stderr).
There's a similar case (with the same message) when the client asks for
an object which we _do_ have, but which isn't a ref tip (or isn't
reachable, when uploadpack.allowReachableSHA1InWant is true). Let's give
that one the same treatment, for the same reason (namely that it's more
informative to the client than just hanging up, since they won't see our
stderr over some protocols).
There are two tests here. We cover it most directly in t5530 by invoking
upload-pack, which matches the existing "not our ref" test.
But a more end-to-end check is that "git fetch" actually shows the
message to the client. We're already checking in t5516 that this case
fails, so we can just check stderr there, too. Note that even after we
started ignoring SIGPIPE in 8bf4becf0c, this could in theory still be
racy as described in that commit (because we die() on write failures
before pumping the connection for any ERR packets).
In practice this should be OK for this case. The server will not
actually check reachability until it has received our whole group of
"want" lines. And since we have no objects in the repository, we won't
send any "have" lines, meaning we're always waiting to read the server
response.
Note also that this case cannot happen in the v2 protocol, since it
allows any available object to be requested. However, we don't have to
take any steps to protect against the upcoming GIT_TEST_PROTOCOL_VERSION
in our tests:
- the tests in t5516 would already need to be skipped under v2, and
that is covered by ab0c5f5096 (tests: always test fetch of
unreachable with v0, 2019-02-25)
- the tests in t5530 invoke upload-pack directly, which will continue
to default to v0. Eventually we may have a test setting which uses
v2 even for bare upload-pack calls, but we can't override it here
until we know what the setting looks like.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git fetch" and "git upload-pack" learned to send all exchange over
the sideband channel while talking the v2 protocol.
* jt/fetch-v2-sideband:
tests: define GIT_TEST_SIDEBAND_ALL
{fetch,upload}-pack: sideband v2 fetch response
sideband: reverse its dependency on pkt-line
pkt-line: introduce struct packet_writer
pack-protocol.txt: accept error packets in any context
Use packet_reader instead of packet_read_line
Update the protocol message specification to allow only the limited
use of scaled quantities. This is ensure potential compatibility
issues will not go out of hand.
* js/filter-options-should-use-plain-int:
filter-options: expand scaled numbers
tree:<depth>: skip some trees even when collecting omits
list-objects-filter: teach tree:# how to handle >0
Define a GIT_TEST_SIDEBAND_ALL environment variable meant to be used
from tests. When set to true, this overrides uploadpack.allowsidebandall
to true, allowing the entire test suite to be run as if this
configuration is in place for all repositories.
As of this patch, all tests pass whether GIT_TEST_SIDEBAND_ALL is unset
or set to 1.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Currently, a response to a fetch request has sideband support only while
the packfile is being sent, meaning that the server cannot send notices
until the start of the packfile.
Extend sideband support in protocol v2 fetch responses to the whole
response. upload-pack will advertise it if the
uploadpack.allowsidebandall configuration variable is set, and
fetch-pack will automatically request it if advertised.
If the sideband is to be used throughout the whole response, upload-pack
will use it to send errors instead of prefixing a PKT-LINE payload with
"ERR ".
This will be tested in a subsequent patch.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When communicating with a remote server or a subprocess, use
expanded numbers rather than numbers with scaling suffix in the
object filter spec (e.g. "limit:blob=1k" becomes
"limit:blob=1024").
Update the protocol docs to note that clients should always perform this
expansion, to allow for more compatibility between server
implementations.
Signed-off-by: Josh Steadmon <steadmon@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A future patch will allow the client to request multiplexing of the
entire fetch response (and not only during packfile transmission), which
in turn allows the server to send progress and keepalive messages at any
time during the response.
It will be convenient for a future patch if writing options
(specifically, whether the written data is to be multiplexed) could be
controlled from a single place, so create struct packet_writer to serve
as that place, and modify upload-pack to use it.
Currently, it only stores the output fd, but a subsequent patch will (as
described above) introduce an option to determine if the written data is
to be multiplexed.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Commit 685fbd3291 ("fetch-pack: perform a fetch using v2", 2018-03-15)
attempted to teach Git deepen-relative in protocol v2 (among other
things), but it didn't work:
(1) fetch-pack.c needs to emit "deepen-relative".
(2) upload-pack.c needs to ensure that the correct deepen_relative
variable is passed to deepen() (there are two - the static variable
and the one in struct upload_pack_data).
(3) Before deepen() computes the list of reachable shallows, it first
needs to mark all "our" refs as OUR_REF. v2 currently does not do
this, because unlike v0, it is not needed otherwise.
Fix all this and include a test demonstrating that it works now. For
(2), the static variable deepen_relative is also eliminated, removing a
source of confusion.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Reviewed-by: Josh Steadmon <steadmon@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the Git pack protocol definition, an error packet may appear only in
a certain context. However, servers can face a runtime error (e.g. I/O
error) at an arbitrary timing. This patch changes the protocol to allow
an error packet to be sent instead of any packet.
Without this protocol spec change, when a server cannot process a
request, there's no way to tell that to a client. Since the server
cannot produce a valid response, it would be forced to cut a connection
without telling why. With this protocol spec change, the server can be
more gentle in this situation. An old client may see these error packets
as an unexpected packet, but this is not worse than having an unexpected
EOF.
Following this protocol spec change, the error packet handling code is
moved to pkt-line.c. Implementation wise, this implementation uses
pkt-line to communicate with a subprocess. Since this is not a part of
Git protocol, it's possible that a packet that is not supposed to be an
error packet is mistakenly parsed as an error packet. This error packet
handling is enabled only for the Git pack protocol parsing code
considering this.
Signed-off-by: Masaya Suzuki <masayasuzuki@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
By using and sharing a packet_reader while handling a Git pack protocol
request, the same reader option is used throughout the code. This makes
it easy to set a reader option to the request parsing code.
Signed-off-by: Masaya Suzuki <masayasuzuki@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git fetch" over protocol v2 into a shallow repository failed to
fetch full history behind a new tip of history that was diverged
before the cut-off point of the history that was previously fetched
shallowly.
* jt/upload-pack-v2-fix-shallow:
upload-pack: clear flags before each v2 request
upload-pack: make want_obj not global
upload-pack: make have_obj not global
More codepaths are moving away from hardcoded hash sizes.
* bc/hash-transition-part-15:
rerere: convert to use the_hash_algo
submodule: make zero-oid comparison hash function agnostic
apply: rename new_sha1_prefix and old_sha1_prefix
apply: replace hard-coded constants
tag: express constant in terms of the_hash_algo
transport: use parse_oid_hex instead of a constant
upload-pack: express constants in terms of the_hash_algo
refs/packed-backend: express constants using the_hash_algo
packfile: express constants in terms of the_hash_algo
pack-revindex: express constants in terms of the_hash_algo
builtin/fetch-pack: remove constants with parse_oid_hex
builtin/mktree: remove hard-coded constant
builtin/repack: replace hard-coded constants
pack-bitmap-write: use GIT_MAX_RAWSZ for allocation
object_id.cocci: match only expressions of type 'struct object_id'
The upload_pack_config() callback uses an if/else chain
like:
if (!strcmp(var, "a"))
...
else if (!strcmp(var, "b"))
...
etc
This works as long as the conditions are mutually exclusive,
but one of them is not. 20b20a22f8 (upload-pack: provide a
hook for running pack-objects, 2016-05-18) added:
else if (current_config_scope() != CONFIG_SCOPE_REPO) {
... check some more options ...
}
That was fine in that commit, because it came at the end of
the chain. But later, 10ac85c785 (upload-pack: add object
filtering for partial clone, 2017-12-08) did this:
else if (current_config_scope() != CONFIG_SCOPE_REPO) {
... check some more options ...
} else if (!strcmp("uploadpack.allowfilter", var))
...
We'd always check the scope condition first, meaning we'd
_only_ respect allowfilter when it's in the repo config. You
can see this with:
git -c uploadpack.allowfilter=true upload-pack . | head -1
which will not advertise the filter capability (but will
after this patch). We never noticed because:
- our tests always set it in the repo config
- in protocol v2, we use a different code path that
actually calls repo_config_get_bool() separately, so
that _does_ work. Real-world people experimenting with
this may be using v2.
The more recent uploadpack.allowrefinwant option is in the
same boat.
There are a few possible fixes:
1. Bump the scope conditional back to the bottom of the
chain. But that just means somebody else is likely to
make the same mistake later.
2. Make the conditional more like the others. I.e.:
else if (!current_config_scope() != CONFIG_SCOPE_REPO &&
!strcmp(var, "uploadpack.notallowedinrepo"))
This works, but the idea of the original structure was
that we may grow multiple sensitive options like this.
3. Pull it out of the chain entirely. The chain mostly
serves to avoid extra strcmp() calls after we've found
a match. But it's not worth caring about those. In the
worst case, when there isn't a match, we're already
hitting every strcmp (and this happens regularly for
stuff like "core.bare", etc).
This patch does (3).
Signed-off-by: Jeff King <peff@peff.net>
Reviewed-by: Josh Steadmon <steadmon@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Suppose a server has the following commit graph:
A B
\ /
O
We create a client by cloning A from the server with depth 1, and add
many commits to it (so that future fetches span multiple requests due to
lengthy negotiation). If it then fetches B using protocol v2, the fetch
spanning multiple requests, the resulting packfile does not contain O
even though the client did report that A is shallow.
This is because upload_pack_v2() can be called multiple times while
processing the same session. During the 2nd and all subsequent
invocations, some object flags remain from the previous invocations. In
particular, CLIENT_SHALLOW remains, preventing process_shallow() from
adding client-reported shallows to the "shallows" array, and hence
pack-objects not knowing about these client-reported shallows.
Therefore, teach upload_pack_v2() to clear object flags at the start of
each invocation. This has some other results:
- THEY_HAVE gates addition of objects to have_obj in process_haves().
Previously in upload_pack_v2(), have_obj needed to be static because
once an object is added to have_obj, it is never readded and thus we
needed to retain the contents of have_obj between invocations. Now
that flags are cleared, this is no longer necessary. This patch does
not change the behavior of ok_to_give_up() (THEY_HAVE is still set on
each "have") and got_oid() (used only in non-v2)); THEY_HAVE is not
used in any other function.
- WANTED gates addition of objects to want_obj in parse_want() and
parse_want_ref(). It is also used in receive_needs(), but that is
only used in non-v2. For the same reasons as THEY_HAVE, want_obj no
longer needs to be static in upload_pack_v2().
- CLIENT_SHALLOW is changed as discussed above.
Clearing of the other 5 flags does not affect functionality in v2. (Note
that in non-v2, upload_pack() is only called once per process, so each
invocation starts with blank flags anyway.)
- OUR_REF is only used in non-v2.
- COMMON_KNOWN is only used as a scratch flag in ok_to_give_up().
- SHALLOW is passed to invocations in deepen() and
deepen_by_rev_list(), but upload-pack doesn't use it.
- NOT_SHALLOW is used by send_shallow() and send_unshallow(), but
invocations of those functions are always preceded by code that sets
NOT_SHALLOW on the appropriate objects.
- HIDDEN_REF is only used in non-v2.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Because upload_pack_v2() can be invoked multiple times in the same
process, the static variable want_obj may not be empty when it is
invoked. To make further analysis of this situation easier, make the
variable local; analysis will be done in a subsequent patch.
The new local variable in upload_pack_v2() is static to preserve
existing behavior; this is not necessary in upload_pack() because
upload_pack() is only invoked once per process.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Because upload_pack_v2() can be invoked multiple times in the same
process, the static variable have_obj may not be empty when it is
invoked. To make further analysis of this situation easier, make the
variable local; analysis will be done in a subsequent patch.
The new local variable in upload_pack_v2() is static to preserve
existing behavior; this is not necessary in upload_pack() because
upload_pack() is only invoked once per process.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The recently introduced commit-graph auxiliary data is incompatible
with mechanisms such as replace & grafts that "breaks" immutable
nature of the object reference relationship. Disable optimizations
based on its use (and updating existing commit-graph) when these
incompatible features are in use in the repository.
* ds/commit-graph-with-grafts:
commit-graph: close_commit_graph before shallow walk
commit-graph: not compatible with uninitialized repo
commit-graph: not compatible with grafts
commit-graph: not compatible with replace objects
test-repository: properly init repo
commit-graph: update design document
refs.c: upgrade for_each_replace_ref to be a each_repo_ref_fn callback
refs.c: migrate internal ref iteration to pass thru repository argument
Convert all uses of the GIT_SHA1_HEXSZ to use the_hash_algo so that they
are appropriate for any given hash length.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The code for computing history reachability has been shuffled,
obtained a bunch of new tests to cover them, and then being
improved.
* ds/reachable:
commit-reach: correct accidental #include of C file
commit-reach: use can_all_from_reach
commit-reach: make can_all_from_reach... linear
commit-reach: replace ref_newer logic
test-reach: test commit_contains
test-reach: test can_all_from_reach_with_flags
test-reach: test reduce_heads
test-reach: test get_merge_bases_many
test-reach: test is_descendant_of
test-reach: test in_merge_bases
test-reach: create new test tool for ref_newer
commit-reach: move can_all_from_reach_with_flags
upload-pack: generalize commit date cutoff
upload-pack: refactor ok_to_give_up()
upload-pack: make reachable() more generic
commit-reach: move commit_contains from ref-filter
commit-reach: move ref_newer from remote.c
commit.h: remove method declarations
commit-reach: move walk methods from commit.c
Call close_commit_graph() when about to start a rev-list walk that
includes shallow commits. This is necessary in code paths that "fake"
shallow commits for the sake of fetch. Specifically, test 351 in
t5500-fetch-pack.sh runs
git fetch --shallow-exclude one origin
with a file-based transfer. When the "remote" has a commit-graph, we do
not prevent the commit-graph from being loaded, but then the commits are
intended to be dynamically transferred into shallow commits during
get_shallow_commits_by_rev_list(). By closing the commit-graph before
this call, we prevent this interaction.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git fetch" failed to correctly validate the set of objects it
received when making a shallow history deeper, which has been
corrected.
* jt/connectivity-check-after-unshallow:
fetch-pack: write shallow, then check connectivity
fetch-pack: implement ref-in-want
fetch-pack: put shallow info in output parameter
fetch: refactor to make function args narrower
fetch: refactor fetch_refs into two functions
fetch: refactor the population of peer ref OIDs
upload-pack: test negotiation with changing repository
upload-pack: implement ref-in-want
test-pkt-line: add unpack-sideband subcommand
The can_all_from_reach_with_flags() algorithm is currently quadratic in
the worst case, because it calls the reachable() method for every 'from'
without tracking which commits have already been walked or which can
already reach a commit in 'to'.
Rewrite the algorithm to walk each commit a constant number of times.
We also add some optimizations that should work for the main consumer of
this method: fetch negotitation (haves/wants).
The first step includes using a depth-first-search (DFS) from each
'from' commit, sorted by ascending generation number. We do not walk
beyond the minimum generation number or the minimum commit date. This
DFS is likely to be faster than the existing reachable() method because
we expect previous ref values to be along the first-parent history.
If we find a target commit, then we mark everything in the DFS stack as
a RESULT. This expands the set of targets for the other 'from' commits.
We also mark the visited commits using 'assign_flag' to prevent re-
walking the same commits.
We still need to clear our flags at the end, which is why we will have a
total of three visits to each commit.
Performance was measured on the Linux repository using
'test-tool reach can_all_from_reach'. The input included rows seeded by
tag values. The "small" case included X-rows as v4.[0-9]* and Y-rows as
v3.[0-9]*. This mimics a (very large) fetch that says "I have all major
v3 releases and want all major v4 releases." The "large" case included
X-rows as "v4.*" and Y-rows as "v3.*". This adds all release-candidate
tags to the set, which does not greatly increase the number of objects
that are considered, but does increase the number of 'from' commits,
demonstrating the quadratic nature of the previous code.
Small Case:
Before: 1.52 s
After: 0.26 s
Large Case:
Before: 3.50 s
After: 0.27 s
Note how the time increases between the two cases in the two versions.
The new code increases relative to the number of commits that need to be
walked, but not directly relative to the number of 'from' commits.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are several commit walks in the codebase. Group them together into
a new commit-reach.c file and corresponding header. After we group these
walks into one place, we can reduce duplicate logic by calling
equivalent methods.
The can_all_from_reach_with_flags method is used in a stateful way by
upload-pack.c. The parameters are very flexible, so we will be able to
use its commit walking logic for many other callers.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The ok_to_give_up() method uses the commit date as a cutoff to avoid
walking the entire reachble set of commits. Before moving the
reachable() method to commit-reach.c, pull out the dependence on the
global constant 'oldest_have' with a 'min_commit_date' parameter.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In anticipation of consolidating all commit reachability algorithms,
refactor ok_to_give_up() in order to allow splitting its logic into
an external method.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
