If you run "GIT_TRACE_PACKET=1 git push", you may get confusing output like (line prefixes omitted for clarity): packet: push< \1000eunpack ok0019ok refs/heads/master0000 packet: push< unpack ok packet: push< ok refs/heads/master packet: push< 0000 packet: push< 0000 Why do we see the data twice, once apparently wrapped inside another pkt-line, and once unwrapped? Why do we get two flush packets? The answer is that we start an async process to demux the sideband data. The first entry comes from the sideband process reading the data, and the second from push itself. Likewise, the first flush is inside the demuxed packet, and the second is an actual sideband flush. We can make this a bit more clear by marking the sideband demuxer explicitly as "sideband" rather than "push". The most elegant way to do this would be to simply call packet_trace_identity() inside the sideband demuxer. But we can't do that reliably, because it relies on a global variable, which might be shared if pthreads are in use. What we really need is thread-local storage for packet_trace_identity. But the async code does not provide an interface for that, and it would be messy to add it here (we'd have to care about pthreads, initializing our pthread_key_t ahead of time, etc). So instead, let us just assume that any async process is handling sideband data. That's always true now, and is likely to remain so in the future. The output looks like: packet: sideband< \1000eunpack ok0019ok refs/heads/master0000 packet: push< unpack ok packet: push< ok refs/heads/master packet: push< 0000 packet: sideband< 0000 Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When debugging the pack protocol, it is sometimes useful to
store the verbatim pack that we sent or received on the
wire. Looking at the on-disk result is often not helpful for
a few reasons:
1. If the operation is a clone, we destroy the repo on
failure, leaving nothing on disk.
2. If the pack is small, we unpack it immediately, and the
full pack never hits the disk.
3. If we feed the pack to "index-pack --fix-thin", the
resulting pack has the extra delta bases added to it.
We already have a GIT_TRACE_PACKET mechanism for tracing
packets. Let's extend it with GIT_TRACE_PACKFILE to dump the
verbatim packfile.
There are a few other positive fallouts that come from
rearranging this code:
- We currently disable the packet trace after seeing the
PACK header, even though we may get human-readable lines
on other sidebands; now we include them in the trace.
- We currently try to print "PACK ..." in the trace to
indicate that the packfile has started. But because we
disable packet tracing, we never printed this line. We
will now do so.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To find the start of the pack data, we accept the word PACK at the beginning of any sideband channel, even though what we really want is to find the pack data on channel 1. In practice this doesn't matter, as sideband-2 messages tend to start with "error:" or similar, but it is a good idea to be explicit (especially as we add more code in this area, we will rely on this assumption). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We carefully check that our pkt buffer has enough characters
before seeing if it starts with "PACK". The intent is to
avoid reading random memory if we get a short buffer like
"PAC".
However, we know that the traced packets are always
NUL-terminated. They come from one of these sources:
1. A string literal.
2. `format_packet`, which uses a strbuf.
3. `packet_read`, which defensively NUL-terminates what we
read.
We can therefore drop the length checks, as we know we will
hit the trailing NUL if we have a short input.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we send out pkt-lines with refnames, we use a static 1000-byte buffer. This means that the maximum size of a ref over the git protocol is around 950 bytes (the exact size depends on the protocol line being written, but figure on a sha1 plus some boilerplate). This is enough for any sane workflow, but occasionally odd things happen (e.g., a bug may create a ref "foo/foo/foo/..." accidentally). With the current code, you cannot even use "push" to delete such a ref from a remote. Let's switch to using a strbuf, with a hard-limit of LARGE_PACKET_MAX (which is specified by the protocol). This matches the size of the readers, as of 74543a0 (pkt-line: provide a LARGE_PACKET_MAX static buffer, 2013-02-20). Versions of git older than that will complain about our large packets, but it's really no worse than the current behavior. Right now the sender barfs with "impossibly long line" trying to send the packet, and afterwards the reader will barf with "protocol error: bad line length %d", which is arguably better anyway. Note that we're not really _solving_ the problem here, but just bumping the limits. In theory, the length of a ref is unbounded, and pkt-line can only represent sizes up to 65531 bytes. So we are just bumping the limit, not removing it. But hopefully 64K should be enough for anyone. As a bonus, by using a strbuf for the formatting we can eliminate an unnecessary copy in format_buf_write. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The trace API currently rechecks the environment variable and reopens the trace file on every API call. This has the ugly side effect that errors (e.g. file cannot be opened, or the user specified a relative path) are also reported on every call. Performance can be improved by about factor three by remembering the environment state and keeping the file open. Replace the 'const char *key' parameter in the API with a pointer to a 'struct trace_key' that bundles the environment variable name with additional, trace-internal state. Change the call sites of these APIs to use a static 'struct trace_key' instead of a string constant. In trace.c::get_trace_fd(), save and reuse the file descriptor in 'struct trace_key'. Add a 'trace_disable()' API, so that packet_trace() can cleanly disable tracing when it encounters packed data (instead of using unsetenv()). Signed-off-by: Karsten Blees <blees@dcon.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Leaving only the function definitions and declarations so that any
new topic in flight can still make use of the old functions, replace
existing uses of the prefixcmp() and suffixcmp() with new API
functions.
The change can be recreated by mechanically applying this:
$ git grep -l -e prefixcmp -e suffixcmp -- \*.c |
grep -v strbuf\\.c |
xargs perl -pi -e '
s|!prefixcmp\(|starts_with\(|g;
s|prefixcmp\(|!starts_with\(|g;
s|!suffixcmp\(|ends_with\(|g;
s|suffixcmp\(|!ends_with\(|g;
'
on the result of preparatory changes in this series.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The packet_read function reads from a descriptor. The packet_get_line function is similar, but reads from an in-memory buffer, and uses a completely separate implementation. This patch teaches the generic packet_read function to accept either source, and we can do away with packet_get_line's implementation. There are two other differences to account for between the old and new functions. The first is that we used to read into a strbuf, but now read into a fixed size buffer. The only two callers are fine with that, and in fact it simplifies their code, since they can use the same static-buffer interface as the rest of the packet_read_line callers (and we provide a similar convenience wrapper for reading from a buffer rather than a descriptor). This is technically an externally-visible behavior change in that we used to accept arbitrary sized packets up to 65532 bytes, and now cap out at LARGE_PACKET_MAX, 65520. In practice this doesn't matter, as we use it only for parsing smart-http headers (of which there is exactly one defined, and it is small and fixed-size). And any extension headers would be breaking the protocol to go over LARGE_PACKET_MAX anyway. The other difference is that packet_get_line would return on error rather than dying. However, both callers of packet_get_line are actually improved by dying. The first caller does its own error checking, but we can drop that; as a result, we'll actually get more specific reporting about protocol breakage when packet_read dies internally. The only downside is that packet_read will not print the smart-http URL that failed, but that's not a big deal; anybody not debugging can already see the remote's URL already, and anybody debugging would want to run with GIT_CURL_VERBOSE anyway to see way more information. The second caller, which is just trying to skip past any extra smart-http headers (of which there are none defined, but which we allow to keep room for future expansion), did not error check at all. As a result, it would treat an error just like a flush packet. The resulting mess would generally cause an error later in get_remote_heads, but now we get error reporting much closer to the source of the problem. Brown-paper-bag-fixes-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk> Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Most of the callers of packet_read_line just read into a
static 1000-byte buffer (callers which handle arbitrary
binary data already use LARGE_PACKET_MAX). This works fine
in practice, because:
1. The only variable-sized data in these lines is a ref
name, and refs tend to be a lot shorter than 1000
characters.
2. When sending ref lines, git-core always limits itself
to 1000 byte packets.
However, the only limit given in the protocol specification
in Documentation/technical/protocol-common.txt is
LARGE_PACKET_MAX; the 1000 byte limit is mentioned only in
pack-protocol.txt, and then only describing what we write,
not as a specific limit for readers.
This patch lets us bump the 1000-byte limit to
LARGE_PACKET_MAX. Even though git-core will never write a
packet where this makes a difference, there are two good
reasons to do this:
1. Other git implementations may have followed
protocol-common.txt and used a larger maximum size. We
don't bump into it in practice because it would involve
very long ref names.
2. We may want to increase the 1000-byte limit one day.
Since packets are transferred before any capabilities,
it's difficult to do this in a backwards-compatible
way. But if we bump the size of buffer the readers can
handle, eventually older versions of git will be
obsolete enough that we can justify bumping the
writers, as well. We don't have plans to do this
anytime soon, but there is no reason not to start the
clock ticking now.
Just bumping all of the reading bufs to LARGE_PACKET_MAX
would waste memory. Instead, since most readers just read
into a temporary buffer anyway, let's provide a single
static buffer that all callers can use. We can further wrap
this detail away by having the packet_read_line wrapper just
use the buffer transparently and return a pointer to the
static storage. That covers most of the cases, and the
remaining ones already read into their own LARGE_PACKET_MAX
buffers.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The packets sent during ref negotiation are all terminated
by newline; even though the code to chomp these newlines is
short, we end up doing it in a lot of places.
This patch teaches packet_read_line to auto-chomp the
trailing newline; this lets us get rid of a lot of inline
chomping code.
As a result, some call-sites which are not reading
line-oriented data (e.g., when reading chunks of packfiles
alongside sideband) transition away from packet_read_line to
the generic packet_read interface. This patch converts all
of the existing callsites.
Since the function signature of packet_read_line does not
change (but its behavior does), there is a possibility of
new callsites being introduced in later commits, silently
introducing an incompatibility. However, since a later
patch in this series will change the signature, such a
commit would have to be merged directly into this commit,
not to the tip of the series; we can therefore ignore the
issue.
This is an internal cleanup and should produce no change of
behavior in the normal case. However, there is one corner
case to note. Callers of packet_read_line have never been
able to tell the difference between a flush packet ("0000")
and an empty packet ("0004"), as both cause packet_read_line
to return a length of 0. Readers treat them identically,
even though Documentation/technical/protocol-common.txt says
we must not; it also says that implementations should not
send an empty pkt-line.
By stripping out the newline before the result gets to the
caller, we will now treat the newline-only packet ("0005\n")
the same as an empty packet, which in turn gets treated like
a flush packet. In practice this doesn't matter, as neither
empty nor newline-only packets are part of git's protocols
(at least not for the line-oriented bits, and readers who
are not expecting line-oriented packets will be calling
packet_read directly, anyway). But even if we do decide to
care about the distinction later, it is orthogonal to this
patch. The right place to tighten would be to stop treating
empty packets as flush packets, and this change does not
make doing so any harder.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>