* jk/tighten-alloc: (23 commits)
compat/mingw: brown paper bag fix for 50a6c8e
ewah: convert to REALLOC_ARRAY, etc
convert ewah/bitmap code to use xmalloc
diff_populate_gitlink: use a strbuf
transport_anonymize_url: use xstrfmt
git-compat-util: drop mempcpy compat code
sequencer: simplify memory allocation of get_message
test-path-utils: fix normalize_path_copy output buffer size
fetch-pack: simplify add_sought_entry
fast-import: simplify allocation in start_packfile
write_untracked_extension: use FLEX_ALLOC helper
prepare_{git,shell}_cmd: use argv_array
use st_add and st_mult for allocation size computation
convert trivial cases to FLEX_ARRAY macros
use xmallocz to avoid size arithmetic
convert trivial cases to ALLOC_ARRAY
convert manual allocations to argv_array
argv-array: add detach function
add helpers for allocating flex-array structs
harden REALLOC_ARRAY and xcalloc against size_t overflow
...
The memory ownership rule of fill_textconv() API, which was a bit
tricky, has been documented a bit better.
* jk/more-comments-on-textconv:
diff: clarify textconv interface
"git merge-tree" used to mishandle "both sides added" conflict with
its own "create a fake ancestor file that has the common parts of
what both sides have added and do a 3-way merge" logic; this has
been updated to use the usual "3-way merge with an empty blob as
the fake common ancestor file" approach used in the rest of the
system.
* jk/no-diff-emit-common:
xdiff: drop XDL_EMIT_COMMON
merge-tree: drop generate_common strategy
merge-one-file: use empty blob for add/add base
The "v(iew)" subcommand of the interactive "git am -i" command was
broken in 2.6.0 timeframe when the command was rewritten in C.
* jc/am-i-v-fix:
am -i: fix "v"iew
pager: factor out a helper to prepare a child process to run the pager
pager: lose a separate argv[]
"git rev-parse --git-common-dir" used in the worktree feature
misbehaved when run from a subdirectory.
* nd/git-common-dir-fix:
rev-parse: take prefix into account in --git-common-dir
"git show 'HEAD:Foo[BAR]Baz'" did not interpret the argument as a
rev, i.e. the object named by the the pathname with wildcard
characters in a tree object.
* nd/dwim-wildcards-as-pathspecs:
get_sha1: don't die() on bogus search strings
check_filename: tighten dwim-wildcard ambiguity
checkout: reorder check_filename conditional
Handling of errors while writing into our internal asynchronous
process has been made more robust, which reduces flakiness in our
tests.
* jk/epipe-in-async:
t5504: handle expected output from SIGPIPE death
test_must_fail: report number of unexpected signal
fetch-pack: ignore SIGPIPE in sideband demuxer
write_or_die: handle EPIPE in async threads
Many codepaths forget to check return value from git_config_set();
the function is made to die() to make sure we do not proceed when
setting a configuration variable failed.
* ps/config-error:
config: rename git_config_set_or_die to git_config_set
config: rename git_config_set to git_config_set_gently
compat: die when unable to set core.precomposeunicode
sequencer: die on config error when saving replay opts
init-db: die on config errors when initializing empty repo
clone: die on config error in cmd_clone
remote: die on config error when manipulating remotes
remote: die on config error when setting/adding branches
remote: die on config error when setting URL
submodule--helper: die on config error when cloning module
submodule: die on config error when linking modules
branch: die on config error when editing branch description
branch: die on config error when unsetting upstream
branch: report errors in tracking branch setup
config: introduce set_or_die wrappers
Traditionally, the tests that try commands that work on the
contents in the working tree were named with "worktree" in their
filenames, but with the recent addition of "git worktree"
subcommand, whose tests are also named similarly, it has become
harder to tell them apart. The traditional tests have been renamed
to use "work-tree" instead in an attempt to differentiate them.
* mg/work-tree-tests:
tests: rename work-tree tests to *work-tree*
Commit 50a6c8e (use st_add and st_mult for allocation size
computation, 2016-02-22) fixed up many xmalloc call-sites
including ones in compat/mingw.c.
But I screwed up one of them, which was half-converted to
ALLOC_ARRAY, using a very early prototype of the function.
And I never caught it because I don't build on Windows.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Commit 8bf4bec (add "ok=sigpipe" to test_must_fail and use
it to fix flaky tests, 2015-11-27) taught t5504 to handle
"git push" racily exiting with SIGPIPE rather than failing.
However, one of the tests checks the output of the command,
as well. In the SIGPIPE case, we will not have produced any
output. If we want the test to be truly non-flaky, we have
to accept either output.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a command is marked as test_must_fail but dies with a
signal, we consider that a problem and report the error to
stderr. However, we don't say _which_ signal; knowing that
can make debugging easier. Let's share as much as we know.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If the other side feeds us a bogus pack, index-pack (or
unpack-objects) may die early, before consuming all of its
input. As a result, the sideband demuxer may get SIGPIPE
(racily, depending on whether our data made it into the pipe
buffer or not). If this happens and we are compiled with
pthread support, it will take down the main thread, too.
This isn't the end of the world, as the main process will
just die() anyway when it sees index-pack failed. But it
does mean we don't get a chance to say "fatal: index-pack
failed" or similar. And it also means that we racily fail
t5504, as we sometimes die() and sometimes are killed by
SIGPIPE.
So let's ignore SIGPIPE while demuxing the sideband. We are
already careful to check the return value of write(), so we
won't waste time writing to a broken pipe. The caller will
notice the error return from the async thread, though in
practice we don't even get that far, as we die() as soon as
we see that index-pack failed.
The non-sideband case is already fine; we let index-pack
read straight from the socket, so there is no SIGPIPE at
all. Technically the non-threaded async case is also OK
without this (the forked async process gets SIGPIPE), but
it's not worth distinguishing from the threaded case here.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When write_or_die() sees EPIPE, it treats it specially by
converting it into a SIGPIPE death. We obviously cannot
ignore it, as the write has failed and the caller expects us
to die. But likewise, we cannot just call die(), because
printing any message at all would be a nuisance during
normal operations.
However, this is a problem if write_or_die() is called from
a thread. Our raised signal ends up killing the whole
process, when logically we just need to kill the thread
(after all, if we are ignoring SIGPIPE, there is good reason
to think that the main thread is expecting to handle it).
Inside an async thread, the die() code already does the
right thing, because we use our custom die_async() routine,
which calls pthread_join(). So ideally we would piggy-back
on that, and simply call:
die_quietly_with_code(141);
or similar. But refactoring the die code to do this is
surprisingly non-trivial. The die_routines themselves handle
both printing and the decision of the exit code. Every one
of them would have to be modified to take new parameters for
the code, and to tell us to be quiet.
Instead, we can just teach write_or_die() to check for the
async case and handle it specially. We do have to build an
interface to abstract the async exit, but it's simple and
self-contained. If we had many call-sites that wanted to do
this die_quietly_with_code(), this approach wouldn't scale
as well, but we don't. This is the only place where do this
weird exit trick.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are no more callers that use this mode, and none
likely to be added (as our xdl_merge() eliminates the common
use of it for generating 3-way merge bases).
This is effectively a revert of a9ed376 (xdiff: generate
"anti-diffs" aka what is common to two files, 2006-06-28),
though of course trying to revert that ancient commit
directly produces many textual conflicts.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When merge_blobs sees an add/add conflict, it tries to
create a virtual base object for the 3-way merge that
consists of the common lines of each file. It inherited this
strategy from merge-one-file in 0c79938 (Improved three-way
blob merging code, 2006-06-28), and the point is to minimize
the size of the conflict hunks. That commit talks about "if
libxdiff were to ever grow a compatible three-way merge, it
could probably be directly plugged in".
That has long since happened. So as with merge-one-file in
the previous commit, this extra step is no longer necessary.
Our 3-way merge code is smart enough to do the minimizing
itself if we simply feed it an empty base, which is what the
more modern merge-recursive strategy already does.
Not only does this let us drop some code, but it removes an
overflow bug in generate_common_file(). We allocate a buffer
as large as the smallest of the two blobs, under the
assumption that there cannot be more common content than
what is in the smaller blob. However, xdiff may feed us
more: if neither file ends in a newline, it feeds us the
"\nNo newline at end of file" marker as common content, and
we write it into the output. If the differences between the
files are small than that string, we overflow the output
buffer. This patch solves it by simply dropping the buggy
code entirely.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we see an add/add conflict on a file, we generate the
conflicted content by doing a 3-way merge with a "virtual"
base consisting of the common lines of the two sides. This
strategy dates back to cb93c19 (merge-one-file: use common
as base, instead of emptiness., 2005-11-09).
Back then, the next step was to call rcs merge to generate
the 3-way conflicts. Using the virtual base produced much
better results, as rcs merge does not attempt to minimize
the hunks. As a result, you'd get a conflict with the
entirety of the files on either side.
Since then, though, we've switched to using git-merge-file,
which uses xdiff's "zealous" merge. This will find the
minimal hunks even with just the simple, empty base.
Let's switch to using that empty base. It's simpler, more
efficient, and reduces our dependencies (we no longer need a
working diff binary). It's also how the merge-recursive
strategy handles this same case.
We can almost get rid of git-sh-setup's create_virtual_base,
but we don't here, for two reasons:
1. The functions in git-sh-setup are part of our public
interface, so it's possible somebody is depending on
it. We'd at least need to deprecate it first.
2. It's also used by mergetool's p4merge driver. It's
unknown whether its 3-way merge is as capable as git's;
if not, then it is benefiting from the function.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now that we're built around xmalloc and friends, we can use
helpers like REALLOC_ARRAY, ALLOC_GROW, and so on to make
the code shorter and protect against integer overflow.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This code was originally written with the idea that it could
be spun off into its own ewah library, and uses the
overrideable ewah_malloc to do allocations.
We plug in xmalloc as our ewah_malloc, of course. But over
the years the ewah code itself has become more entangled
with git, and the return value of many ewah_malloc sites is
not checked.
Let's just drop the level of indirection and use xmalloc and
friends directly. This saves a few lines, and will let us
adapt these sites to our more advanced malloc helpers.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We allocate 100 bytes to hold the "Submodule commit ..."
text. This is enough, but it's not immediately obvious that
this is the case, and we have to repeat the magic 100 twice.
We could get away with xstrfmt here, but we want to know the
size, as well, so let's use a real strbuf. And while we're
here, we can clean up the logic around size_only. It
currently sets and clears the "data" field pointlessly, and
leaves the "should_free" flag on even after we have cleared
the data.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This function uses xcalloc and two memcpy calls to
concatenate two strings. We can do this as an xstrfmt
one-liner, and then it is more clear that we are allocating
the correct amount of memory.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are no callers of this left, as the last one was
dropped in the previous patch. And there are not likely to
be new ones, as the function has been around since 2010
without gaining any new callers.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
For a commit with sha1 "1234abcd" and subject "foo", this
function produces a struct with three strings:
1. "foo"
2. "1234abcd... foo"
3. "parent of 1234abcd... foo"
It takes advantage of the fact that these strings are
subsets of each other, and allocates only _one_ string, with
pointers into the various parts. Unfortunately, this makes
the string allocation complicated and hard to follow.
Since we keep only one of these in memory at a time, we can
afford to simply allocate three strings. This lets us build
on tools like xstrfmt and avoid manual computation.
While we're here, we can also drop the ad-hoc
reimplementation of get_git_commit_encoding(), and simply
call that function.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The normalize_path_copy function needs an output buffer that
is at least as long as its input (it may shrink the path,
but never expand it). However, this test program feeds it
static PATH_MAX-sized buffers, which have no relation to the
input size.
In the normalize_ceiling_entry case, we do at least check
the size against PATH_MAX and die(), but that case is even
more convoluted. We normalize into a fixed-size buffer, free
the original, and then replace it with a strdup'd copy of
the result. But normalize_path_copy explicitly allows
normalizing in-place, so we can simply do that.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We have two variants of this function, one that takes a
string and one that takes a ptr/len combo. But we only call
the latter with the length of a NUL-terminated string, so
our first simplification is to drop it in favor of the
string variant.
Since we know we have a string, we can also replace the
manual memory computation with a call to alloc_ref().
Furthermore, we can rely on get_oid_hex() to complain if it
hits the end of the string. That means we can simplify the
check for "<sha1> <ref>" versus just "<ref>". Rather than
manage the ptr/len pair, we can just bump the start of our
string forward. The original code over-allocated based on
the original "namelen" (which wasn't _wrong_, but was simply
wasteful and confusing).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This function allocate a packed_git flex-array, and adds a
mysterious 2 bytes to the length of the pack_name field. One
is for the trailing NUL, but the other has no purpose. This
is probably cargo-culted from add_packed_git, which gets the
".idx" path and needed to allocate enough space to hold the
matching ".pack" (though since 48bcc1c, we calculate the
size there differently).
This site, however, is using the raw path of a tempfile, and
does not need the extra byte. We can just replace the
allocation with FLEX_ALLOC_STR, which handles the allocation
and the NUL for us.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We perform unchecked additions when computing the size of a
"struct ondisk_untracked_cache". This is unlikely to have an
integer overflow in practice, but we'd like to avoid this
dangerous pattern to make further audits easier.
Note that there's one subtlety here, though. We protect
ourselves against a NULL exclude_per_dir entry in our
source, and avoid calling strlen() on it, keeping "len" at
0. But later, we unconditionally memcpy "len + 1" bytes to
get the trailing NUL byte. If we did have a NULL
exclude_per_dir, we would read from bogus memory.
As it turns out, though, we always create this field
pointing to a string literal, so there's no bug. We can just
get rid of the pointless extra conditional.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
These functions transform an existing argv into one suitable
for exec-ing or spawning via git or a shell. We can use an
argv_array in each to avoid dealing with manual counting and
allocation.
This also makes the memory allocation more clear and fixes
some leaks. In prepare_shell_cmd, we would sometimes
allocate a new string with "$@" in it and sometimes not,
meaning the caller could not correctly free it. On the
non-Windows side, we are in a child process which will
exec() or exit() immediately, so the leak isn't a big deal.
On Windows, though, we use spawn() from the parent process,
and leak a string for each shell command we run. On top of
that, the Windows code did not free the allocated argv array
at all (but does for the prepare_git_cmd case!).
By switching both of these functions to write into an
argv_array, we can consistently free the result as
appropriate.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If our size computation overflows size_t, we may allocate a
much smaller buffer than we expected and overflow it. It's
probably impossible to trigger an overflow in most of these
sites in practice, but it is easy enough convert their
additions and multiplications into overflow-checking
variants. This may be fixing real bugs, and it makes
auditing the code easier.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Using FLEX_ARRAY macros reduces the amount of manual
computation size we have to do. It also ensures we don't
overflow size_t, and it makes sure we write the same number
of bytes that we allocated.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We frequently allocate strings as xmalloc(len + 1), where
the extra 1 is for the NUL terminator. This can be done more
simply with xmallocz, which also checks for integer
overflow.
There's no case where switching xmalloc(n+1) to xmallocz(n)
is wrong; the result is the same length, and malloc made no
guarantees about what was in the buffer anyway. But in some
cases, we can stop manually placing NUL at the end of the
allocated buffer. But that's only safe if it's clear that
the contents will always fill the buffer.
In each case where this patch does so, I manually examined
the control flow, and I tried to err on the side of caution.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Each of these cases can be converted to use ALLOC_ARRAY or
REALLOC_ARRAY, which has two advantages:
1. It automatically checks the array-size multiplication
for overflow.
2. It always uses sizeof(*array) for the element-size,
so that it can never go out of sync with the declared
type of the array.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are many manual argv allocations that predate the
argv_array API. Switching to that API brings a few
advantages:
1. We no longer have to manually compute the correct final
array size (so it's one less thing we can screw up).
2. In many cases we had to make a separate pass to count,
then allocate, then fill in the array. Now we can do it
in one pass, making the code shorter and easier to
follow.
3. argv_array handles memory ownership for us, making it
more obvious when things should be free()d and and when
not.
Most of these cases are pretty straightforward. In some, we
switch from "run_command_v" to "run_command" which lets us
directly use the argv_array embedded in "struct
child_process".
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The usual pattern for an argv array is to initialize it,
push in some strings, and then clear it when done. Very
occasionally, though, we must do other exotic things with
the memory, like freeing the list but keeping the strings.
Let's provide a detach function so that callers can make use
of our API to build up the array, and then take ownership of
it.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Allocating a struct with a flex array is pretty simple in
practice: you over-allocate the struct, then copy some data
into the over-allocation. But it can be a slight pain to
make sure you're allocating and copying the right amounts.
This patch adds a few helpers to turn simple cases of
flex-array struct allocation into a one-liner that properly
checks for overflow. See the embedded documentation for
details.
Ideally we could provide a more flexible version that could
handle multiple strings, like:
FLEX_ALLOC_FMT(ref, name, "%s%s", prefix, name);
But we have to implement this as a macro (because of the
offset calculation of the flex member), which means we would
need all compilers to support variadic macros.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
REALLOC_ARRAY inherently involves a multiplication which can
overflow size_t, resulting in a much smaller buffer than we
think we've allocated. We can easily harden it by using
st_mult() to check for overflow. Likewise, we can add
ALLOC_ARRAY to do the same thing for xmalloc calls.
xcalloc() should already be fine, because it takes the two
factors separately, assuming the system calloc actually
checks for overflow. However, before we even hit the system
calloc(), we do our memory_limit_check, which involves a
multiplication. Let's check for overflow ourselves so that
this limit cannot be bypassed.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Paths that have been told the index about with "add -N" are not
quite yet in the index, but a few commands behaved as if they
already are in a harmful way.
* nd/ita-cleanup:
grep: make it clear i-t-a entries are ignored
add and use a convenience macro ce_intent_to_add()
blame: remove obsolete comment
The documentation for "git clean" has been corrected; it mentioned
that .git/modules/* are removed by giving two "-f", which has never
been the case.
* mm/clean-doc-fix:
Documentation/git-clean.txt: don't mention deletion of .git/modules/*
The vimdiff backend for "git mergetool" has been tweaked to arrange
and number buffers in the order that would match the expectation of
majority of people who read left to right, then top down and assign
buffers 1 2 3 4 "mentally" to local base remote merge windows based
on that order.
* dw/mergetool-vim-window-shuffle:
mergetool: reorder vim/gvim buffers in three-way diffs
The memory allocation scheme for the textconv interface is a
bit tricky, and not well documented. It was originally
designed as an internal part of diff.c (matching
fill_mmfile), but gradually was made public.
Refactoring it is difficult, but we can at least improve the
situation by documenting the intended flow and enforcing it
with an in-code assertion.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Rename git_config_set_or_die functions to git_config_set, leading
to the new default behavior of dying whenever a configuration
error occurs.
By now all callers that shall die on error have been transitioned
to the _or_die variants, thus making this patch a simple rename
of the functions.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The desired default behavior for `git_config_set` is to die
whenever an error occurs. Dying is the default for a lot of
internal functions when failures occur and is in this case the
right thing to do for most callers as otherwise we might run into
inconsistent repositories without noticing.
As some code may rely on the actual return values for
`git_config_set` we still require the ability to invoke these
functions without aborting. Rename the existing `git_config_set`
functions to `git_config_set_gently` to keep them available for
those callers.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When calling `git_config_set` to set 'core.precomposeunicode' we
ignore the return value of the function, which may indicate that
we were unable to write the value back to disk. As the function
is only called by init-db we can and should die when an error
occurs.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we start picking a range of revisions we save the replay
options that are required to restore state when interrupting and
later continuing picking the revisions. However, we do not check
the return values of the `git_config_set` functions, which may
lead us to store incomplete information. As this may lead us to
fail when trying to continue the sequence the error can be fatal.
Fix this by dying immediately when we are unable to write back
any replay option.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When creating an empty repository with `git init-db` we do not
check for error codes returned by `git_config_set` functions.
This may cause the user to end up with an inconsistent repository
without any indication for the user.
Fix this problem by dying early with an error message when we are
unable to write the configuration files to disk.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
