The infrastructure to rewrite "git submodule" in C is being built
incrementally. Let's polish these early parts well enough and make
them graduate to 'next' and 'master', so that the more involved
follow-up can start cooking on a solid ground.
* sb/submodule-helper:
submodule: rewrite `module_clone` shell function in C
submodule: rewrite `module_name` shell function in C
submodule: rewrite `module_list` shell function in C
The "ref-filter" code was taught about many parts of what "tag -l"
does and then "tag -l" is being reimplemented in terms of "ref-filter".
* kn/for-each-tag:
tag.c: implement '--merged' and '--no-merged' options
tag.c: implement '--format' option
tag.c: use 'ref-filter' APIs
tag.c: use 'ref-filter' data structures
ref-filter: add option to match literal pattern
ref-filter: add support to sort by version
ref-filter: add support for %(contents:lines=X)
ref-filter: add option to filter out tags, branches and remotes
ref-filter: implement an `align` atom
ref-filter: introduce match_atom_name()
ref-filter: introduce handler function for each atom
utf8: add function to align a string into given strbuf
ref-filter: introduce ref_formatting_state and ref_formatting_stack
ref-filter: move `struct atom_value` to ref-filter.c
strtoul_ui: reject negative values
Because "test_when_finished" in our test framework queues the
clean-up tasks to be done in a shell variable, it should not be
used inside a subshell. Add a mechanism to allow 'bash' to catch
such uses, and fix the ones that were found.
* jk/test-lint-forbid-when-finished-in-subshell:
test-lib-functions: detect test_when_finished in subshell
t7800: don't use test_config in a subshell
test-lib-functions: support "test_config -C <dir> ..."
t5801: don't use test_when_finished in a subshell
t7610: don't use test_config in a subshell
Update "git subtree" (in contrib/) so that it can take whitespaces
in the pathnames, not only in the in-tree pathname but the name of
the directory that the repository is in.
* as/subtree-with-spaces:
contrib/subtree: respect spaces in a repository path
t7900-subtree: test the "space in a subdirectory name" case
The ssh transport, just like any other transport over the network,
did not clear GIT_* environment variables, but it is possible to
use SendEnv and AcceptEnv to leak them to the remote invocation of
Git, which is not a good idea at all. Explicitly clear them just
like we do for the local transport.
* jk/connect-clear-env:
git_connect: clarify conn->use_shell flag
git_connect: clear GIT_* environment for ssh
"git log --date=local" used to only show the normal (default)
format in the local timezone. The command learned to take 'local'
as an instruction to use the local timezone with other formats,
e.g. "git show --date=rfc-local".
* jk/date-local:
t6300: add tests for "-local" date formats
t6300: make UTC and local dates different
date: make "local" orthogonal to date format
date: check for "local" before anything else
t6300: add test for "raw" date format
t6300: introduce test_date() helper
fast-import: switch crash-report date to iso8601
Documentation/rev-list: don't list date formats
Documentation/git-for-each-ref: don't list date formats
Documentation/config: don't list date formats
Documentation/blame-options: don't list date formats
Move the refs used during a "git bisect" session to per-worktree
hierarchy refs/worktree/* so that independent bisect sessions can
be done in different worktrees.
* dt/refs-bisection:
refs: make refs/bisect/* per-worktree
path: optimize common dir checking
refs: clean up common_list
Users who are too busy to type three extra keystrokes to ask for
"git stash show -p" can now set stash.showPatch configuration
varible to true to always see the actual patch, not just the list
of paths affected with feel for the extent of damage via diffstat.
Correct "git p4 --detect-labels" so that it does not fail to create
a tag that points at a commit that is also being imported.
* ld/p4-import-labels:
git-p4: fix P4 label import for unprocessed commits
git-p4: do not terminate creating tag for unknown commit
git-p4: failing test for ignoring invalid p4 labels
The use of 'good/bad' in "git bisect" made it confusing to use when
hunting for a state change that is not a regression (e.g. bugfix).
The command learned 'old/new' and then allows the end user to
say e.g. "bisect start --term-old=fast --term=new=slow" to find a
performance regression.
Michael's idea to make 'good/bad' more intelligent does have
certain attractiveness ($gname/272867), and makes some of the work
on this topic a moot point.
* ad/bisect-terms:
bisect: allow setting any user-specified in 'git bisect start'
bisect: add 'git bisect terms' to view the current terms
bisect: add the terms old/new
bisect: sanity check on terms
* jc/rerere: (21 commits)
rerere: un-nest merge() further
rerere: use "struct rerere_id" instead of "char *" for conflict ID
rerere: call conflict-ids IDs
rerere: further clarify do_rerere_one_path()
rerere: further de-dent do_plain_rerere()
rerere: refactor "replay" part of do_plain_rerere()
rerere: explain the remainder
rerere: explain "rerere forget" codepath
rerere: explain the primary codepath
rerere: explain MERGE_RR management helpers
rerere: fix benign off-by-one non-bug and clarify code
rerere: explain the rerere I/O abstraction
rerere: do not leak mmfile[] for a path with multiple stage #1 entries
rerere: stop looping unnecessarily
rerere: drop want_sp parameter from is_cmarker()
rerere: report autoupdated paths only after actually updating them
rerere: write out each record of MERGE_RR in one go
rerere: lift PATH_MAX limitation
rerere: plug conflict ID leaks
rerere: handle conflicts with multiple stage #1 entries
...
Some features from "git tag -l" and "git branch -l" have been made
available to "git for-each-ref" so that eventually the unified
implementation can be shared across all three, in a follow-up
series or two.
The manual size computations here are correct, but using
strip_suffix makes that obvious, and hopefully communicates
the intent of the code more clearly.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When working with paths in strbufs, we frequently want to
ensure that a directory contains a trailing slash before
appending to it. We can shorten this code (and make the
intent more obvious) by calling strbuf_complete.
Most of these cases are trivially identical conversions, but
there are two things to note:
- in a few cases we did not check that the strbuf is
non-empty (which would lead to an out-of-bounds memory
access). These were generally not triggerable in
practice, either from earlier assertions, or typically
because we would have just fed the strbuf to opendir(),
which would choke on an empty path.
- in a few cases we indexed the buffer with "original_len"
or similar, rather than the current sb->len, and it is
not immediately obvious from the diff that they are the
same. In all of these cases, I manually verified that
the strbuf does not change between the assignment and
the strbuf_complete call.
This does not convert cases which look like:
if (sb->len && !is_dir_sep(sb->buf[sb->len - 1]))
strbuf_addch(sb, '/');
as those are obviously semantically different. Some of these
cases arguably should be doing that, but that is out of
scope for this change, which aims purely for cleanup with no
behavior change (and at least it will make such sites easier
to find and examine in the future, as we can grep for
strbuf_complete).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 27e1e22 (prune: factor out loose-object directory
traversal, 2014-10-15), we now have a generic callback
system for iterating over the loose object directories. This
is used by prune, count-objects, etc.
We did not convert git-fsck at the time because it
implemented an inode-sorting scheme that was not part of the
generic code. Now that the inode-sorting code is gone, we
can reuse the generic code. The result is shorter,
hopefully more readable, and drops some unchecked sprintf
calls.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fsck tries to access loose objects in order of inode number,
with the hope that this would make cold cache access faster
on a spinning disk. This dates back to 7e8c174 (fsck-cache:
sort entries by inode number, 2005-05-02), which predates
the invention of packfiles.
These days, there's not much point in trying to optimize
cold cache for a large number of loose objects. You are much
better off to simply pack the objects, which will reduce the
disk footprint _and_ provide better locality of data access.
So while you can certainly construct pathological cases
where this code might help, it is not worth the trouble
anymore.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
strncpy is known to be a confusing function because of its
termination semantics. These calls are all correct, but it
takes some examination to see why. In particular, every one
of them expects to copy up to the length limit, and then
makes some arrangement for terminating the result.
We can just use memcpy, along with noting explicitly how the
result is terminated (if it is not already obvious). That
should make it more clear to a reader that we are doing the
right thing.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
notes: document length of fanout path with a constant
We know that a fanned-out sha1 in a notes tree cannot be
more than "aa/bb/cc/...", and we have an assert() to confirm
that. But let's factor out that length into a constant so we
can be sure it is used consistently. And even though we
assert() earlier, let's replace a strcpy with xsnprintf, so
it is clear to a reader that all cases are covered.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
color: add color_set helper for copying raw colors
To set up default colors, we sometimes strcpy() from the
default string literals into our color buffers. This isn't a
bug (assuming the destination is COLOR_MAXLEN bytes), but
makes it harder to audit the code for problematic strcpy
calls.
Let's introduce a color_set which copies under the
assumption that there are COLOR_MAXLEN bytes in the
destination (of course you can call it on a smaller buffer,
so this isn't providing a huge amount of safety, but it's
more convenient than calling xsnprintf yourself).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we already know the length of a string (e.g., because
we just malloc'd to fit it), it's nicer to use memcpy than
strcpy, as it makes it more obvious that we are not going to
overflow the buffer (because the size we pass matches the
size in the allocation).
This also eliminates calls to strcpy, which make auditing
the code base harder.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we are going to launch "/path/to/konqueror", we instead
rewrite this into "/path/to/kfmclient" by duplicating the
original string and writing over the ending bits. This can
be done more obviously with strip_suffix and xstrfmt.
Note that we also fix a subtle bug with the "filename"
parameter, which is passed as argv[0] to the child. If the
user has configured a program name with no directory
component, we always pass the string "kfmclient", even if
your program is called something else. But if you give a
full path, we give the basename of that path. But more
bizarrely, if we rewrite "konqueror" to "kfmclient", we
still pass "konqueror".
The history of this function doesn't reveal anything
interesting, so it looks like just an oversight from
combining the suffix-munging with the basename-finding.
Let's just call basename on the munged path, which produces
consistent results (if you gave a program, whether a full
path or not, we pass its basename).
Probably this doesn't matter at all in practice, but it
makes the code slightly less confusing to read.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
To generate "--keep=receive-pack $pid on $host", we write
progressively into a single buffer, which requires keeping
track of how much we've written so far. But since the result
is destined to go into our argv array, we can simply use
argv_array_pushf.
Unfortunately we still have to have a fixed-size buffer for
the gethostname() call, but at least it now doesn't involve
any extra size computation. And as a bonus, we drop an
sprintf and a strcpy call.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we are allocating a struct with a FLEX_ARRAY member, we
generally compute the size of the array and then sprintf or
strcpy into it. Normally we could improve a dynamic allocation
like this by using xstrfmt, but it doesn't work here; we
have to account for the size of the rest of the struct.
But we can improve things a bit by storing the length that
we use for the allocation, and then feeding it to xsnprintf
or memcpy, which makes it more obvious that we are not
writing more than the allocated number of bytes.
It would be nice if we had some kind of helper for
allocating generic flex arrays, but it doesn't work that
well:
- the call signature is a little bit unwieldy:
d = flex_struct(sizeof(*d), offsetof(d, path), fmt, ...);
You need offsetof here instead of just writing to the
end of the base size, because we don't know how the
struct is packed (partially this is because FLEX_ARRAY
might not be zero, though we can account for that; but
the size of the struct may actually be rounded up for
alignment, and we can't know that).
- some sites do clever things, like over-allocating because
they know they will write larger things into the buffer
later (e.g., struct packed_git here).
So we're better off to just write out each allocation (or
add type-specific helpers, though many of these are one-off
allocations anyway).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our color parsing is designed to never exceed COLOR_MAXLEN
bytes. But the relationship between that hand-computed
number and the parsing code is not at all obvious, and we
merely hope that it has been computed correctly for all
cases.
Let's mark the expected "end" pointer for the destination
buffer and make sure that we do not exceed it.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
In some cases where we strcpy() the result of sha1_to_hex(),
there's no need; the result goes directly into a printf
statement, and we can simply pass the return value from
sha1_to_hex() directly.
When this code was originally written, sha1_to_hex used a
single buffer, and it was not safe to use it twice within a
single expression. That changed as of dcb3450 (sha1_to_hex()
usage cleanup, 2006-05-03), but this code was never updated.
History-dug-by: Eric Sunshine <sunshine@sunshineco.com> Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before sha1_to_hex_r() existed, a simple way to get hex
sha1 into a buffer was with:
strcpy(buf, sha1_to_hex(sha1));
This isn't wrong (assuming the buf is 41 characters), but it
makes auditing the code base for bad strcpy() calls harder,
as these become false positives.
Let's convert them to sha1_to_hex_r(), and likewise for
some calls to find_unique_abbrev(). While we're here, we'll
double-check that all of the buffers are correctly sized,
and use the more obvious GIT_SHA1_HEXSZ constant.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This avoids an ugly strcat into a fixed-size buffer. It's
not wrong (the buffer is plenty large enough for an IPv6
address plus some minor formatting), but it takes some
effort to verify that.
Unfortunately we are still stuck with some fixed-size
buffers to hold the output of inet_ntop. But at least we now
pass very easy-to-verify parameters, rather than doing a
manual computation to account for other data in the buffer.
As a side effect, this also fixes the case where we might
pass an uninitialized portbuf buffer through the
environment. This probably couldn't happen in practice, as
it would mean that addr->sa_family was neither AF_INET nor
AF_INET6 (and that is all we are listening on).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This drops the magic number for the fixed-size argv arrays,
so we do not have to wonder if we are overflowing it. We can
also drop some confusing sha1_to_hex memory allocation
(which seems to predate the ring of buffers allowing
multiple calls), and get rid of an unchecked sprintf call.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
fetch-pack: use argv_array for index-pack / unpack-objects
This cleans up a magic number that must be kept in sync with
the rest of the code (the number of argv slots). It also
lets us drop some fixed buffers and an sprintf (since we
can now use argv_array_pushf).
We do still have to keep one fixed buffer for calling
gethostname, but at least now the size computations for it
are much simpler.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
diagnose_invalid_index_path: use strbuf to avoid strcpy/strcat
We dynamically allocate a buffer and then strcpy and strcat
into it. This isn't buggy, but we'd prefer to avoid these
suspicious functions.
This would be a good candidate for converstion to xstrfmt,
but we need to record the length for dealing with index
entries. A strbuf handles that for us.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When creating a loose object tempfile, we use a fixed
PATH_MAX-sized buffer, and strcpy directly into it. This
isn't buggy, because we do a rough check of the size, but
there's no verification that our guesstimate of the required
space is enough (in fact, it's several bytes too big for the
current naming scheme).
Let's switch to a strbuf, which makes this much easier to
verify. The allocation overhead should be negligible, since
we are replacing a static buffer with a static strbuf, and
we'll only need to allocate on the first call.
While we're here, we can also document a subtle interaction
with mkstemp that would be easy to overlook.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
remove_leading_path: use a strbuf for internal storage
This function strcpy's directly into a PATH_MAX-sized
buffer. There's only one caller, which feeds the git_dir into
it, so it's not easy to trigger in practice (even if you fed
a large $GIT_DIR through the environment or .git file, it
would have to actually exist and be accessible on the
filesystem to get to this point). We can fix it by moving to
a strbuf.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We use two PATH_MAX-sized buffers to represent the repo
path, and must make sure not to overflow them. We do take
care to check the lengths, but the logic is rather hard to
follow, as we use several magic numbers (e.g., "PATH_MAX -
10"). And in fact you _can_ overflow the buffer if you have
a ".git" file with an extremely long path in it.
By switching to strbufs, these problems all go away. We do,
however, retain the check that the initial input we get is
no larger than PATH_MAX. This function is an entry point for
untrusted repo names from the network, and it's a good idea
to keep a sanity check (both to avoid allocating arbitrary
amounts of memory, and also as a layer of defense against
any downstream users of the names).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
merge-recursive: convert malloc / strcpy to strbuf
This would be a fairly routine use of xstrfmt, except that
we need to remember the length of the result to pass to
cache_name_pos. So just use a strbuf, which makes this
simple.
As a bonus, this gets rid of confusing references to
"pathlen+1". The "1" is for the trailing slash we added, but
that is automatically accounted for in the strbuf's len
parameter.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
transport: use strbufs for status table "quickref" strings
We generate range strings like "1234abcd...5678efab" for use
in the the fetch and push status tables. We use fixed-size
buffers along with strcat to do so. These aren't buggy, as
our manual size computation is correct, but there's nothing
checking that this is so. Let's switch them to strbufs
instead, which are obviously correct, and make it easier to
audit the code base for problematic calls to strcat().
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We use manual computation and strcpy to allocate the "root"
variable. This would be much simpler using xstrfmt. But
since we store the length, too, we can just use a strbuf,
which handles that for us.
Note that we stop distinguishing between "no root" and
"empty root" in some cases, but that's OK; the results are
the same (e.g., inserting an empty string is a noop).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The init code predates strbufs, and uses PATH_MAX-sized
buffers along with many manual checks on intermediate sizes
(some of which make magic assumptions, such as that init
will not create a path inside .git longer than 50
characters).
We can simplify this greatly by using strbufs, which drops
some hard-to-verify strcpy calls in favor of git_path_buf.
While we're in the area, let's also convert existing calls
to git_path to the safer git_path_buf (our existing calls
were passed to pretty tame functions, and so were not a
problem, but it's easy to be consistent and safe here).
Note that we had an explicit test that "git init" rejects
long template directories. This comes from 32d1776 (init: Do
not segfault on big GIT_TEMPLATE_DIR environment variable,
2009-04-18). We can drop the test_must_fail here, as we now
accept this and need only confirm that we don't segfault,
which was the original point of the test.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
probe_utf8_pathname_composition: use internal strbuf
When we are initializing a .git directory, we may call
probe_utf8_pathname_composition to detect utf8 mangling. We
pass in a path buffer for it to use, and it blindly
strcpy()s into it, not knowing whether the buffer is large
enough to hold the result or not.
In practice this isn't a big deal, because the buffer we
pass in already contains "$GIT_DIR/config", and we append
only a few extra bytes to it. But we can easily do the right
thing just by calling git_path_buf ourselves. Technically
this results in a different pathname (before we appended our
utf8 characters to the "config" path, and now they get their
own files in $GIT_DIR), but that should not matter for our
purposes.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The result of iconv is assigned to a variable, but we never
use it (instead, we check errno and whether the function
consumed all bytes). Let's drop the assignment, as it
triggers gcc's -Wunused-but-set-variable.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Perforce repositories can contain large (binary) files. Migrating these
repositories to Git generates very large local clones. External storage
systems such as Git LFS [1], Git Fat [2], Git Media [3], git-annex [4]
try to address this problem.
Add a generic mechanism to detect large files based on extension,
uncompressed size, and/or compressed size.
Signed-off-by: Lars Schneider <larsxschneider@gmail.com>
Conflicts:
Documentation/git-p4.txt
git-p4.py Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-p4 will just halt if there is not enough disk space while
streaming content from P4 to Git. Add a check to ensure at least
4 times (arbitrarily chosen) the size of a streamed file is available.
Signed-off-by: Lars Schneider <larsxschneider@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-p4: add optional type specifier to gitConfig reader
The functions "gitConfig" and "gitConfigBool" are almost identical.
Make "gitConfig" more generic by adding an optional type specifier.
Use the type specifier "--bool" with "gitConfig" to implement
"gitConfigBool. This prepares the implementation of other type
specifiers such as "--int".
Signed-off-by: Lars Schneider <larsxschneider@gmail.com> Acked-by: Luke Diamand <luke@diamand.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
rebase-i: explicitly accept tab as separator in commands
The git-rebase-todo is parsed several times with different parsers. In
principle, the user input is normalized by transform_todo_ids and
further parsing can be stricter.
the parser of transform_todo_ids was considering the sha1 to be
"deadbeef<TAB>commit", and was leaving the tab in the transformed sheet.
In practice, this went unnoticed since the actual command interpretation
was done later in do_next which did accept the tab as a separator.
Make it explicit in the code of transform_todo_ids that tabs are
accepted. This way, code that mimicks it will also accept tabs as
separator.
A similar construct appears in skip_unnecessary_picks, but this one
comes after transform_todo_ids, hence reads the normalized format, so it
needs not be changed.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
After using "git checkout --detach", the reflog is left with an entry
like
checkout: moving from ... to HEAD
This message is parsed to generate the 'HEAD detached at' message in
'git branch' and 'git status', which leads to the not-so-useful message
'HEAD detached at HEAD'.
Instead, when parsing such reflog entry, resolve HEAD to the
corresponding commit in the reflog, so that the message becomes 'HEAD
detached at $sha1'.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The new builtin am ignores the user.signingkey variable: gpg is being
called with the committer details as the key ID, which may not be
correct. git_gpg_config is responsible for handling that variable and is
expected to be called on initialization by any modules that use gpg.
Signed-off-by: Renee Margaret McConahy <nepella@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-send-email.perl: Fixed sending of many/huge changes/patches
Sometimes sending huge patches/commits fail with
[Net::SMTP::SSL] Connection closed at /usr/lib/git-core/git-send-email
line 1320.
Running the command with --smtp-debug=1 yields to
Net::SMTP::SSL: Net::Cmd::datasend(): unexpected EOF on command channel:
at /usr/lib/git-core/git-send-email line 1320.
[Net::SMTP::SSL] Connection closed at /usr/lib/git-core/git-send-email
line 1320.
Stefan described it in his mail like this:
It seems to me that there is a size limit, after cutting down the patch
to ~16K, sending started to work. I cut it twice, once by removing lines
from the head and once from the bottom, in both cases at the size of
around 16K I could send the patch.
See also original report:
http://permalink.gmane.org/gmane.comp.version-control.git/274569
Reported-by: Juston Li <juston.h.li@gmail.com> Tested-by: Markos Chandras <hwoarang@gentoo.org> Signed-off-by: Lars Wendler <polynomial-c@gentoo.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
merge-file: enforce MAX_XDIFF_SIZE on incoming files
The previous commit enforces MAX_XDIFF_SIZE at the
interfaces to xdiff: xdi_diff (which calls xdl_diff) and
ll_xdl_merge (which calls xdl_merge).
But we have another direct call to xdl_merge in
merge-file.c. If it were written today, this probably would
just use the ll_merge machinery. But it predates that code,
and uses slightly different options to xdl_merge (e.g.,
ZEALOUS_ALNUM).
We could try to abstract out an xdi_merge to match the
existing xdi_diff, but even that is difficult. Rather than
simply report error, we try to treat large files as binary,
and that distinction would happen outside of xdi_merge.
The simplest fix is to just replicate the MAX_XDIFF_SIZE
check in merge-file.c.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The xdiff code is not prepared to handle extremely large
files. It uses "int" in many places, which can overflow if
we have a very large number of lines or even bytes in our
input files. This can cause us to produce incorrect diffs,
with no indication that the output is wrong. Or worse, we
may even underallocate a buffer whose size is the result of
an overflowing addition.
We're much better off to tell the user that we cannot diff
or merge such a large file. This patch covers both cases,
but in slightly different ways:
1. For merging, we notice the large file and cleanly fall
back to a binary merge (which is effectively "we cannot
merge this").
2. For diffing, we make the binary/text distinction much
earlier, and in many different places. For this case,
we'll use the xdi_diff as our choke point, and reject
any diff there before it hits the xdiff code.
This means in most cases we'll die() immediately after.
That's not ideal, but in practice we shouldn't
generally hit this code path unless the user is trying
to do something tricky. We already consider files
larger than core.bigfilethreshold to be binary, so this
code would only kick in when that is circumvented
(either by bumping that value, or by using a
.gitattribute to mark a file as diffable).
In other words, we can avoid being "nice" here, because
there is already nice code that tries to do the right
thing. We are adding the suspenders to the nice code's
belt, so notice when it has been worked around (both to
protect the user from malicious inputs, and because it
is better to die() than generate bogus output).
The maximum size was chosen after experimenting with feeding
large files to the xdiff code. It's just under a gigabyte,
which leaves room for two obvious cases:
- a diff3 merge conflict result on files of maximum size X
could be 3*X plus the size of the markers, which would
still be only about 3G, which fits in a 32-bit int.
- some of the diff code allocates arrays of one int per
record. Even if each file consists only of blank lines,
then a file smaller than 1G will have fewer than 1G
records, and therefore the int array will fit in 4G.
Since the limit is arbitrary anyway, I chose to go under a
gigabyte, to leave a safety margin (e.g., we would not want
to overflow by allocating "(records + 1) * sizeof(int)" or
similar.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we call into xdiff to perform a diff, we generally lose
the return code completely. Typically by ignoring the return
of our xdi_diff wrapper, but sometimes we even propagate
that return value up and then ignore it later. This can
lead to us silently producing incorrect diffs (e.g., "git
log" might produce no output at all, not even a diff header,
for a content-level diff).
In practice this does not happen very often, because the
typical reason for xdiff to report failure is that it
malloc() failed (it uses straight malloc, and not our
xmalloc wrapper). But it could also happen when xdiff
triggers one our callbacks, which returns an error (e.g.,
outf() in builtin/rerere.c tries to report a write failure
in this way). And the next patch also plans to add more
failure modes.
Let's notice an error return from xdiff and react
appropriately. In most of the diff.c code, we can simply
die(), which matches the surrounding code (e.g., that is
what we do if we fail to load a file for diffing in the
first place). This is not that elegant, but we are probably
better off dying to let the user know there was a problem,
rather than simply generating bogus output.
We could also just die() directly in xdi_diff, but the
callers typically have a bit more context, and can provide a
better message (and if we do later decide to pass errors up,
we're one step closer to doing so).
There is one interesting case, which is in diff_grep(). Here
if we cannot generate the diff, there is nothing to match,
and we silently return "no hits". This is actually what the
existing code does already, but we make it a little more
explicit.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
-u <exec> has never been supported, but it was mentioned since 0a2bb55 (git ls-remote: make usage string match manpage -
2008-11-11). Nobody has complained about it for seven years, it's
probably safe to say nobody cares. So let's remove "-u" in documents
instead of adding code to support it.
While at there, fix --upload-pack syntax too.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
setup: fix "inside work tree" detection on case-insensitive filesystems
Git has a config variable to indicate that it is operating on a file
system that is case-insensitive: core.ignoreCase. But the
`dir_inside_of()` function did not respect that. As a result, if Git's
idea of the current working directory disagreed in its upper/lower case
with the `GIT_WORK_TREE` variable (e.g. `C:\test` vs `c:\test`) the
user would be greeted by the error message
fatal: git-am cannot be used without a working tree.
when trying to run a rebase.
This fixes https://github.com/git-for-windows/git/issues/402 (reported by
Daniel Harding).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Strict mode is about not guessing where .git is. If the user points to a
.git file, we know exactly where the target .git dir will be. This makes
it possible to serve .git files as repository on the server side.
This may be needed even in local clone case because transport.c code
uses upload-pack for fetching remote refs. But right now the
clone/transport code goes with non-strict.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
enter_repo: avoid duplicating logic, use is_git_directory() instead
It matters for linked checkouts where 'refs' directory won't be
available in $GIT_DIR. is_git_directory() knows about $GIT_COMMON_DIR
and can handle this case.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
By default, libcurl will follow circular http redirects
forever. Let's put a cap on this so that somebody who can
trigger an automated fetch of an arbitrary repository (e.g.,
for CI) cannot convince git to loop infinitely.
The value chosen is 20, which is the same default that
Firefox uses.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Previously, libcurl would follow redirection to any protocol
it was compiled for support with. This is desirable to allow
redirection from HTTP to HTTPS. However, it would even
successfully allow redirection from HTTP to SFTP, a protocol
that git does not otherwise support at all. Furthermore
git's new protocol-whitelisting could be bypassed by
following a redirect within the remote helper, as it was
only enforced at transport selection time.
This patch limits redirects within libcurl to HTTP, HTTPS,
FTP and FTPS. If there is a protocol-whitelist present, this
list is limited to those also allowed by the whitelist. As
redirection happens from within libcurl, it is impossible
for an HTTP redirect to a protocol implemented within
another remote helper.
When the curl version git was compiled with is too old to
support restrictions on protocol redirection, we warn the
user if GIT_ALLOW_PROTOCOL restrictions were requested. This
is a little inaccurate, as even without that variable in the
environment, we would still restrict SFTP, etc, and we do
not warn in that case. But anything else means we would
literally warn every time git accesses an http remote.
This commit includes a test, but it is not as robust as we
would hope. It redirects an http request to ftp, and checks
that curl complained about the protocol, which means that we
are relying on curl's specific error message to know what
happened. Ideally we would redirect to a working ftp server
and confirm that we can clone without protocol restrictions,
and not with them. But we do not have a portable way of
providing an ftp server, nor any other protocol that curl
supports (https is the closest, but we would have to deal
with certificates).
[jk: added test and version warning]
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Asciidoctor is stricter than AsciiDoc when deciding if underlining
is a section title or the start of preformatted text. Make the
length of the underlining match the text to ensure that it renders
correctly in all implementations.
Signed-off-by: John Keeping <john@keeping.me.uk>
[jc: squashed in git-bisect one noticed by Michael J Gruber] Signed-off-by: Junio C Hamano <gitster@pobox.com>
t/perf: make runner work even if Git is not installed
aggregate.perl did not work when Git.pm is not installed to a directory
contained in the default Perl library path list or PERLLIB.
This commit prepends the Perl library path of the current Git source
tree to enable this.
Note that this commit adds a hard-coded relative path
use lib '../../perl/blib/lib';
instead of the flexible environment-based variant
use lib (split(/:/, $ENV{GITPERLLIB}));
which is used in tests written in Perl.
The hard-coded variant is used because the whole performance test
framework does it that way (and GITPERLLIB is not set there).
Signed-off-by: Stephan Beyer <s-beyer@gmx.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We do some manual memory computation here, and there's no
check that our 60 is not overflowed by the raw sprintf (it
isn't, because the "which" parameter is never longer than
"pack"). We can simplify this greatly with a strbuf.
Technically the end result is not identical, as the original
took care not to rewrite the object directory on each call
for performance reasons. We could do that here, too (by
saving the baselen and resetting to it), but it's not worth
the complexity; this function is not called a lot (generally
once per packfile that we open).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We do an unchecked sprintf directly into our url buffer.
This doesn't overflow because we know that it was sized for
"$base/objects/info/http-alternates", and we are writing
"$base/objects/info/alternates", which must be smaller. But
that is not immediately obvious to a reader who is looking
for buffer overflows. Let's switch to a strbuf, so that we
do not have to think about this issue at all.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The http-push code defines an fwrite_buffer function for use
as a curl callback; it just writes to a strbuf. There's no
reason we need to use it ourselves, as we know we have a
strbuf. This lets us format directly into it, rather than
dealing with an extra temporary buffer (which required
manual length computation).
While we're here, let's also remove the literal tabs from
the source in favor of "\t", which is more visually obvious.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We format a pkt-line into a heap buffer, which requires
manual computation of the required size, and uses some bare
sprintf calls. We could use a strbuf instead, which would
take care of the computation for us. But it's even easier
still to use packet_write(). Besides handling the formatting
and writing for us, it fixes two things:
1. Our manual max-size check used 0xFFFF, while technically
LARGE_PACKET_MAX is slightly smaller than this.
2. Our packet will now be output as part of
GIT_TRACE_PACKET debugging.
Unfortunately packet_write() does not let us build up the
buffer progressively, so we do have to repeat ourselves a
little depending on the "vhost" setting, but the end result
is still far more readable than the original.
Since there were no tests covering this feature at all,
we'll add a few into t5802.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we report an error to the client, we format it into a
fixed-size buffer using vsprintf(). This can't actually
overflow in practice, since we only format a very tame
subset of strings (mostly strerror() output). However, it's
hard to tell immediately, so let's just use a strbuf so
readers do not have to wonder.
We do add an allocation here, but the performance is not
important; the next step is to call die() anyway.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
resolve_ref already uses a strbuf internally when generating
pathnames, but it uses fixed-size buffers for storing the
refname and symbolic refs. This means that you cannot
actually point HEAD to a ref that is larger than 256 bytes.
We can lift this limit by using strbufs here, too. Like
sb_path, we pass the the buffers into our helper function,
so that we can easily clean up all output paths. We can also
drop the "unsafe" name from our helper function, as it no
longer uses a single static buffer (but of course
resolve_ref_unsafe is still unsafe, because the static
buffers moved there).
As a bonus, we also get to drop some strcpy calls between
the two fixed buffers (that cannot currently overflow
because the two buffers are sized identically).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The main motivation for this cleanup is to switch our
line-reading to a strbuf, which removes the use of a
fixed-size buffer (which limited the size of remote URLs).
Since we have the strbuf, we can make use of strbuf_rtrim().
While we're here, we can also simplify the parsing of each
line. First, we can use skip_prefix() to avoid some magic
numbers.
But second, we can avoid splitting the parsing and actions
for each line into two stages. Right now we figure out which
type of line we have, set an int to a magic number,
skip any intermediate whitespace, and then act on
the resulting value based on the magic number.
Instead, let's factor the whitespace skipping into a
function. That lets us avoid the magic numbers and keep the
actions close to the parsing.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This function does a lot of manual string handling, and has
some unnecessary limits. This patch cleans up a number of
things:
1. Drop the arbitrary 1000-byte limit on the size of the
remote name (we do not have such a limit in any of the
other remote-reading mechanisms).
2. Replace fgets into a fixed-size buffer with a strbuf,
eliminating any limits on the length of the URL.
3. Replace manual whitespace handling with strbuf_trim
(since we now have a strbuf). This also gets rid
of a call to strcpy, and the confusing reuse of the "p"
pointer for multiple purposes.
4. We currently build up the refspecs over multiple strbuf
calls. We do this to handle the fact that the URL "frag"
may not be present. But rather than have multiple
conditionals, let's just default "frag" to "master".
This lets us format the refspecs with a single xstrfmt.
It's shorter, and easier to see what the final string
looks like.
We also update the misleading comment in this area (the
local branch is named after the remote name, not after
the branch name on the remote side).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We want to make a copy of a string without any leading
whitespace. To do so, we allocate a buffer large enough to
hold the original, skip past the whitespace, then copy that.
It's much simpler to just allocate after we've skipped, in
which case we can just copy the remainder of the string,
leaving no question of whether "len" is large enough.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The ref-filter code comes from for-each-ref, and inherited a
number of raw sprintf and strcpy calls. These are generally
all safe, as we custom-size the buffers, or are formatting
numbers into sufficiently large buffers. But we can make the
resulting code even simpler and more obviously correct by
using some of our helper functions.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we want to convert "foo.pack" to "foo.idx", we do it by
duplicating the original string and then munging the bytes
in place. Let's use strip_suffix and xstrfmt instead, which
has several advantages:
1. It's more clear what the intent is.
2. It does not implicitly rely on the fact that
strlen(".idx") <= strlen(".pack") to avoid an overflow.
3. We communicate the assumption that the input file ends
with ".pack" (and get a run-time check that this is so).
4. We drop calls to strcpy, which makes auditing the code
base easier.
Likewise, we can do this to convert ".pack" to ".bitmap",
avoiding some manual memory computation.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We parse the INFINITE_DEPTH constant into a static,
fixed-size buffer using sprintf. This buffer is sufficiently
large for the current constant, but it's a suspicious
pattern, as the constant is defined far away, and it's not
immediately obvious that 12 bytes are large enough to hold
it.
We can just use xstrfmt here, which gets rid of any question
of the buffer size. It also removes any concerns with object
lifetime, which means we do not have to wonder why this
buffer deep within a conditional is marked "static" (we
never free our newly allocated result, of course, but that's
OK; it's global that lasts the lifetime of the whole program
anyway).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We xmalloc a fixed-size buffer and sprintf into it; this is
OK because the size of our formatting types is finite, but
that's not immediately clear to a reader auditing sprintf
calls. Let's switch to xstrfmt, which is shorter and
obviously correct.
Note that just dropping the common xmalloc here causes gcc
to complain with -Wmaybe-uninitialized. That's because if
"types" does not match any of our known types, we never
write anything into the "normalized" pointer. With the
current code, gcc doesn't notice because we always return a
valid pointer (just one which might point to uninitialized
data, but the compiler doesn't know that). In other words,
the current code is potentially buggy if new types are added
without updating this spot.
So let's take this opportunity to clean up the function a
bit more. We can drop the "normalized" pointer entirely, and
just return directly from each code path. And then add an
assertion at the end in case we haven't covered any cases.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>