In a pthread-enabled version of upload-pack, there's a race condition
that can cause a deadlock on the fflush(NULL) we call from run-command.
What happens is this:
1. Upload-pack is informed we are doing a shallow clone.
2. We call start_async() to spawn a thread that will generate rev-list
results to feed to pack-objects. It gets a file descriptor to a
pipe which will eventually hook to pack-objects.
3. The rev-list thread uses fdopen to create a new output stream
around the fd we gave it, called pack_pipe.
4. The thread writes results to pack_pipe. Outside of our control,
libc is doing locking on the stream. We keep writing until the OS
pipe buffer is full, and then we block in write(), still holding
the lock.
5. The main thread now uses start_command to spawn pack-objects.
Before forking, it calls fflush(NULL) to flush every stdio output
buffer. It blocks trying to get the lock on pack_pipe.
And we have a deadlock. The thread will block until somebody starts
reading from the pipe. But nobody will read from the pipe until we
finish flushing to the pipe.
To fix this, we swap the start order: we start the
pack-objects reader first, and then the rev-list writer
after. Thus the problematic fflush(NULL) happens before we
even open the new file descriptor (and even if it didn't,
flushing should no longer block, as the reader at the end of
the pipe is now active).
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>
Do not advertise no-done capability when upload-pack is not serving over
smart-http, as there is no way for this server to know when it should stop
reading in-flight data from the client, even though it is necessary to
drain all the in-flight data in order to unblock the client.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
This reverts 3e63b21 (upload-pack: Implement no-done capability,
2011-03-14). Together with 761ecf0 (fetch-pack: Implement no-done
capability, 2011-03-14) it seems to make the fetch-pack process out of
sync and makes it keep talking long after upload-pack stopped listening to
it, terminating the process with SIGPIPE.
If the client requests both multi_ack_detailed and no-done then
upload-pack is free to immediately send a PACK following its first
'ACK %s ready' message. The upload-pack response actually winds
up being:
ACK %s common
... (maybe more) ...
ACK %s ready
NAK
ACK %s
PACK.... the pack stream ....
For smart HTTP connections this saves one HTTP RPC, reducing
the overall latency for a trivial fetch. For git:// and ssh://
a no-done option slightly reduces latency by removing one
server->client->server round-trip at the end of the common
ancestor negotiation.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a client is merely following the remote (and has not made any
new commits itself), all "have %s" lines sent by the client will be
common to the server. As all lines are common upload-pack never
calls ok_to_give_up() and does not compute if it has a good cut
point in the commit graph.
Without this computation the following client is going to send all
tagged commits, as these were determined to be COMMON_REF during the
initial advertisement, but the client does not parse their history
to transitively pass the COMMON flag and empty its queue of commits.
For git.git with 339 commit tags, it takes clients 11 rounds of
negotation to fully send all tagged commits and exhaust its queue
of things to send as common. This is pretty slow for a client that
has not done any local development activity.
Force computing ok_to_give_up() and send "ACK %s ready" at the end
of the current round if this round only contained common objects
and ok_to_give_up() was therefore not called. This may allow the
client to break early, avoiding transmission of the COMMON_REFs.
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>
When printing an error message saying a ref was requested that we do not
have, only print that ref, rather than the ref and everything sent to us
on the same packet line (e.g. protocol support specifications).
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A bit of history in chronological order, the newest at bottom:
- 80ccaa7 (upload-pack: Move the revision walker into a separate function.)
do_rev_list was introduced with create_full_pack argument
- 21edd3f (upload-pack: Run rev-list in an asynchronous function.)
do_rev_list was now called by start_async, create_full_pack was
passed by rev_list.data
- f0cea83 (Shift object enumeration out of upload-pack)
rev_list.data was now zero permanently. Creating full pack was
done by passing --all to pack-objects
- ae6a560 (run-command: support custom fd-set in async)
rev_list.data = 0 was found out redudant and got rid of.
Get rid of the code as well, for less headache while reading do_rev_list.
[jc: noticed by Elijah Newren]
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This patch adds the possibility to supply a set of non-0 file
descriptors for async process communication instead of the
default-created pipe.
Additionally, we now support bi-directional communiction with the
async procedure, by giving the async function both read and write
file descriptors.
To retain compatiblity and similar "API feel" with start_command,
we require start_async callers to set .out = -1 to get a readable
file descriptor. If either of .in or .out is 0, we supply no file
descriptor to the async process.
[sp: Note: Erik started this patch, and a huge bulk of it is
his work. All bugs were introduced later by Shawn.]
Signed-off-by: Erik Faye-Lund <kusmabite@gmail.com>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This hook runs after "git fetch" in the repository the objects are
fetched from as the user who fetched, and has security implications.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* sp/smart-http: (37 commits)
http-backend: Let gcc check the format of more printf-type functions.
http-backend: Fix access beyond end of string.
http-backend: Fix bad treatment of uintmax_t in Content-Length
t5551-http-fetch: Work around broken Accept header in libcurl
t5551-http-fetch: Work around some libcurl versions
http-backend: Protect GIT_PROJECT_ROOT from /../ requests
Git-aware CGI to provide dumb HTTP transport
http-backend: Test configuration options
http-backend: Use http.getanyfile to disable dumb HTTP serving
test smart http fetch and push
http tests: use /dumb/ URL prefix
set httpd port before sourcing lib-httpd
t5540-http-push: remove redundant fetches
Smart HTTP fetch: gzip requests
Smart fetch over HTTP: client side
Smart push over HTTP: client side
Discover refs via smart HTTP server when available
http-backend: more explict LocationMatch
http-backend: add example for gitweb on same URL
http-backend: use mod_alias instead of mod_rewrite
...
Conflicts:
.gitignore
remote-curl.c
In theory it is possible for sideband channel #2 to be delayed if
pack data is quick to come up for sideband channel #1. And because
data for channel #2 is read only 128 bytes at a time while pack data
is read 8192 bytes at a time, it is possible for many pack blocks to
be sent to the client before the progress message fifo is emptied,
making the situation even worse. This would result in totally garbled
progress display on the client's console as local progress gets mixed
with partial remote progress lines.
Let's prevent such situations by giving transmission priority to
progress messages over pack data at all times.
Signed-off-by: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When --stateless-rpc is passed as a command line parameter to
upload-pack or receive-pack the programs now assume they may
perform only a single read-write cycle with stdin and stdout.
This fits with the HTTP POST request processing model where a
program may read the request, write a response, and must exit.
When --advertise-refs is passed as a command line parameter only
the initial ref advertisement is output, and the program exits
immediately. This fits with the HTTP GET request model, where
no request content is received but a response must be produced.
HTTP headers and/or environment are not processed here, but
instead are assumed to be handled by the program invoking
either service backend.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When multi_ack_detailed is enabled the ACK continue messages returned
by the remote upload-pack are broken out to describe the different
states within the peer. This permits the client to better understand
the server's in-memory state.
The fetch-pack/upload-pack protocol now looks like:
NAK
---------------------------------
Always sent in response to "done" if there was no common base
selected from the "have" lines (or no have lines were sent).
* no multi_ack or multi_ack_detailed:
Sent when the client has sent a pkt-line flush ("0000") and
the server has not yet found a common base object.
* either multi_ack or multi_ack_detailed:
Always sent in response to a pkt-line flush.
ACK %s
-----------------------------------
* no multi_ack or multi_ack_detailed:
Sent in response to "have" when the object exists on the remote
side and is therefore an object in common between the peers.
The argument is the SHA-1 of the common object.
* either multi_ack or multi_ack_detailed:
Sent in response to "done" if there are common objects.
The argument is the last SHA-1 determined to be common.
ACK %s continue
-----------------------------------
* multi_ack only:
Sent in response to "have".
The remote side wants the client to consider this object as
common, and immediately stop transmitting additional "have"
lines for objects that are reachable from it. The reason
the client should stop is not given, but is one of the two
cases below available under multi_ack_detailed.
ACK %s common
-----------------------------------
* multi_ack_detailed only:
Sent in response to "have". Both sides have this object.
Like with "ACK %s continue" above the client should stop
sending have lines reachable for objects from the argument.
ACK %s ready
-----------------------------------
* multi_ack_detailed only:
Sent in response to "have".
The client should stop transmitting objects which are reachable
from the argument, and send "done" soon to get the objects.
If the remote side has the specified object, it should
first send an "ACK %s common" message prior to sending
"ACK %s ready".
Clients may still submit additional "have" lines if there are
more side branches for the client to explore that might be added
to the common set and reduce the number of objects to transfer.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There were several unchecked use of fdopen(); replace them with xfdopen()
that checks and dies.
Signed-off-by: Jim Meyering <meyering@redhat.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 2d14d65 (Use a clearer style to issue commands to remote helpers,
2009-09-03) I happened to notice two changes like this:
- write_in_full(helper->in, "list\n", 5);
+
+ strbuf_addstr(&buf, "list\n");
+ write_in_full(helper->in, buf.buf, buf.len);
+ strbuf_reset(&buf);
IMHO, it would be better to define a new function,
static inline ssize_t write_str_in_full(int fd, const char *str)
{
return write_in_full(fd, str, strlen(str));
}
and then use it like this:
- strbuf_addstr(&buf, "list\n");
- write_in_full(helper->in, buf.buf, buf.len);
- strbuf_reset(&buf);
+ write_str_in_full(helper->in, "list\n");
Thus not requiring the added allocation, and still avoiding
the maintenance risk of literal string lengths.
These days, compilers are good enough that strlen("literal")
imposes no run-time cost.
Transformed via this:
perl -pi -e \
's/write_in_full\((.*?), (".*?"), \d+\)/write_str_in_full($1, $2)/'\
$(git grep -l 'write_in_full.*"')
Signed-off-by: Jim Meyering <meyering@redhat.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
First of all, I can't find any reason why thin pack generation is
explicitly disabled when dealing with a shallow repository. The
possible delta base objects are collected from the edge commits which
are always obtained through history walking with the same shallow refs
as the client, Therefore the client is always going to have those base
objects available. So let's remove that restriction.
Then we can make shallow repository deepening much more efficient by
using the remote's unshallowed commits as edge commits to get preferred
base objects for thin pack generation. On git.git, this makes the data
transfer for the deepening of a shallow repository from depth 1 to depth 2
around 134 KB instead of 3.68 MB.
Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The majority of code in core git appears to use a single
space after if/for/while. This is an attempt to bring more
code to this standard. These are entirely cosmetic changes.
Signed-off-by: Brian Gianforcaro <b.gianfo@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A request to clone the repository does not give any "have" but asks for
all the refs we offer with "want". When a request does not ask to clone
the repository fully, but asks to fetch some refs into an empty
repository, it will not give any "have" but its "want" won't ask for all
the refs we offer.
If we suppose (and I would say this is a rather big if) that it makes
sense to distinguish these two cases, a hook cannot reliably do this
alone. The hook can detect lack of "have" and bunch of "want", but there
is no direct way to tell if the other end asked for all refs we offered,
or merely most of them.
Between the time we talked with the other end and the time the hook got
called, we may have acquired more refs or lost some refs in the repository
by concurrent operations. Given that we plan to introduce selective
advertisement of refs with a protocol extension, it would become even more
difficult for hooks to guess between these two cases.
This adds "kind [clone|fetch]" to hook's input, as a stable interface to
allow the hooks to tell these cases apart.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
After upload-pack successfully finishes its operation, post-upload-pack
hook can be called for logging purposes.
The hook is passed various pieces of information, one per line, from its
standard input. Currently the following items can be fed to the hook, but
more types of information may be added in the future:
want SHA-1::
40-byte hexadecimal object name the client asked to include in the
resulting pack. Can occur one or more times in the input.
have SHA-1::
40-byte hexadecimal object name the client asked to exclude from
the resulting pack, claiming to have them already. Can occur zero
or more times in the input.
time float::
Number of seconds spent for creating the packfile.
size decimal::
Size of the resulting packfile in bytes.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* cc/replace:
t6050: check pushing something based on a replaced commit
Documentation: add documentation for "git replace"
Add git-replace to .gitignore
builtin-replace: use "usage_msg_opt" to give better error messages
parse-options: add new function "usage_msg_opt"
builtin-replace: teach "git replace" to actually replace
Add new "git replace" command
environment: add global variable to disable replacement
mktag: call "check_sha1_signature" with the replacement sha1
replace_object: add a test case
object: call "check_sha1_signature" with the replacement sha1
sha1_file: add a "read_sha1_file_repl" function
replace_object: add mechanism to replace objects found in "refs/replace/"
refs: add a "for_each_replace_ref" function
upload-pack runs pack-objects, which generates progress indicator output
on its stderr. If the client requests a sideband, this indicator is sent
to the client; but if it did not, then the progress is written to
upload-pack's own stderr.
If upload-pack is itself run from git-daemon (and if the client did not
request a sideband) the progress indicator never reaches the client and it
need not be generated in the first place. With this patch the progress
indicator is suppressed in this situation.
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Offload object enumeration in upload-pack to pack-objects, but fall
back on internal revision walker for shallow interaction. Aside from
architecturally making more sense, this also leaves the door open for
pack-objects to employ a revision cache mechanism. Test t5530 updated
in order to explicitly check both enumeration methods.
Signed-off-by: Nick Edelen <sirnot@gmail.com>
Acked-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This new "read_replace_refs" global variable is set to 1 by
default, so that replace refs are used by default. But
reachability traversal and packing commands ("cmd_fsck",
"cmd_prune", "cmd_pack_objects", "upload_pack",
"cmd_unpack_objects") set it to 0, as they must work with the
original DAG.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* lt/pack-object-memuse:
show_object(): push path_name() call further down
process_{tree,blob}: show objects without buffering
Conflicts:
builtin-pack-objects.c
builtin-rev-list.c
list-objects.c
list-objects.h
upload-pack.c
In particular, pushing the "path_name()" call _into_ the show() function
would seem to allow
- more clarity into who "owns" the name (ie now when we free the name in
the show_object callback, it's because we generated it ourselves by
calling path_name())
- not calling path_name() at all, either because we don't care about the
name in the first place, or because we are actually happy walking the
linked list of "struct name_path *" and the last component.
Now, I didn't do that latter optimization, because it would require some
more coding, but especially looking at "builtin-pack-objects.c", we really
don't even want the whole pathname, we really would be better off with the
list of path components.
Why? We use that name for two things:
- add_preferred_base_object(), which actually _wants_ to traverse the
path, and now does it by looking for '/' characters!
- for 'name_hash()', which only cares about the last 16 characters of a
name, so again, generating the full name seems to be just unnecessary
work.
Anyway, so I didn't look any closer at those things, but it did convince
me that the "show_object()" calling convention was crazy, and we're
actually better off doing _less_ in list-objects.c, and giving people
access to the internal data structures so that they can decide whether
they want to generate a path-name or not.
This patch does that, and then for people who did use the name (even if
they might do something more clever in the future), it just does the
straightforward "name = path_name(path, component); .. free(name);" thing.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Here's a less trivial thing, and slightly more dubious one.
I was looking at that "struct object_array objects", and wondering why we
do that. I have honestly totally forgotten. Why not just call the "show()"
function as we encounter the objects? Rather than add the objects to the
object_array, and then at the very end going through the array and doing a
'show' on all, just do things more incrementally.
Now, there are possible downsides to this:
- the "buffer using object_array" _can_ in theory result in at least
better I-cache usage (two tight loops rather than one more spread out
one). I don't think this is a real issue, but in theory..
- this _does_ change the order of the objects printed. Instead of doing a
"process_tree(revs, commit->tree, &objects, NULL, "");" in the loop
over the commits (which puts all the root trees _first_ in the object
list, this patch just adds them to the list of pending objects, and
then we'll traverse them in that order (and thus show each root tree
object together with the objects we discover under it)
I _think_ the new ordering actually makes more sense, but the object
ordering is actually a subtle thing when it comes to packing
efficiency, so any change in order is going to have implications for
packing. Good or bad, I dunno.
- There may be some reason why we did it that odd way with the object
array, that I have simply forgotten.
Anyway, now that we don't buffer up the objects before showing them
that may actually result in lower memory usage during that whole
traverse_commit_list() phase.
This is seriously not very deeply tested. It makes sense to me, it seems
to pass all the tests, it looks ok, but...
Does anybody remember why we did that "object_array" thing? It used to be
an "object_list" a long long time ago, but got changed into the array due
to better memory usage patterns (those linked lists of obejcts are
horrible from a memory allocation standpoint). But I wonder why we didn't
do this back then. Maybe there's a reason for it.
Or maybe there _used_ to be a reason, and no longer is.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The goal of this patch is to get rid of the "static struct rev_info
revs" static variable in "builtin-rev-list.c".
To do that, we need to pass the revs to the "show_commit" function
in "builtin-rev-list.c" and this in turn means that the
"traverse_commit_list" function in "list-objects.c" must be passed
functions pointers to functions with 2 parameters instead of one.
So we have to change all the callers and all the functions passed
to "traverse_commit_list".
Anyway this makes the code more clean and more generic, so it
should be a good thing in the long run.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
These weren't used outside and can be safely moved
Signed-off-by: Benjamin Kramer <benny.kra@googlemail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Certain remote commands, when asked to do something in a
particular directory that was not actually a git repository,
would say "unable to chdir or not a git archive". The
"chdir" bit is an unnecessary detail, and the term "git
archive" is much less common these days than "git repository".
So let's switch them all to:
fatal: '%s' does not appear to be a git repository
Signed-off-by: Jeff King <peff@peff.net>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Programs that use git_config need to find the global configuration.
When runtime prefix computation is enabled, this requires that
git_extract_argv0_path() is called early in the program's main().
This commit adds the necessary calls.
Signed-off-by: Steffen Prohaska <prohaska@zib.de>
Acked-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is a mechanical conversion of all '*.c' files with:
s/((?:die|error|warning)\("git)-(\S+:)/$1 $2/;
The result was manually inspected and no false positive was found.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We will need the command invocation path in system_path(). This path was
passed to setup_path(), but system_path() can be called earlier, for
example via:
main
commit_pager_choice
setup_pager
git_config
git_etc_gitconfig
system_path
Therefore, we introduce git_set_argv0_path() and call it as soon as
possible.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In upload-pack we must explicitly close the output channel of rev-list.
(On Unix, the channel is closed automatically because process that runs
rev-list terminates.)
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
The new protocol extension "include-tag" allows the client side
of the connection (fetch-pack) to request that the server side of the
native git protocol (upload-pack / pack-objects) use --include-tag
as it prepares the packfile, thus ensuring that an annotated tag object
will be included in the resulting packfile if the object it refers to
was also included into the packfile.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
