Previous changes introduced a regression which will prevent root for
accessing repositories owned by thyself if using sudo because SUDO_UID
takes precedence.
Loosen that restriction by allowing root to access repositories owned
by both uid by default and without having to add a safe.directory
exception.
A previous workaround that was documented in the tests is no longer
needed so it has been removed together with its specially crafted
prerequisite.
Helped-by: Johanness Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a support library that provides one function that can be used
to run a "scriplet" of commands through sudo and that helps invoking
sudo in the slightly awkward way that is required to ensure it doesn't
block the call (if shell was allowed as tested in the prerequisite)
and it doesn't run the command through a different shell than the one
we intended.
Add additional negative tests as suggested by Junio and that use a
new workspace that is owned by root.
Document a regression that was introduced by previous commits where
root won't be able anymore to access directories they own unless
SUDO_UID is removed from their environment.
The tests document additional ways that this new restriction could
be worked around and the documentation explains why it might be instead
considered a feature, but a "fix" is planned for a future change.
Helped-by: Junio C Hamano <gitster@pobox.com>
Helped-by: Phillip Wood <phillip.wood123@gmail.com>
Signed-off-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
bdc77d1d685 (Add a function to determine whether a path is owned by the
current user, 2022-03-02) checks for the effective uid of the running
process using geteuid() but didn't account for cases where that user was
root (because git was invoked through sudo or a compatible tool) and the
original uid that repository trusted for its config was no longer known,
therefore failing the following otherwise safe call:
guy@renard ~/Software/uncrustify $ sudo git describe --always --dirty
[sudo] password for guy:
fatal: unsafe repository ('/home/guy/Software/uncrustify' is owned by someone else)
Attempt to detect those cases by using the environment variables that
those tools create to keep track of the original user id, and do the
ownership check using that instead.
This assumes the environment the user is running on after going
privileged can't be tampered with, and also adds code to restrict that
the new behavior only applies if running as root, therefore keeping the
most common case, which runs unprivileged, from changing, but because of
that, it will miss cases where sudo (or an equivalent) was used to change
to another unprivileged user or where the equivalent tool used to raise
privileges didn't track the original id in a sudo compatible way.
Because of compatibility with sudo, the code assumes that uid_t is an
unsigned integer type (which is not required by the standard) but is used
that way in their codebase to generate SUDO_UID. In systems where uid_t
is signed, sudo might be also patched to NOT be unsigned and that might
be able to trigger an edge case and a bug (as described in the code), but
it is considered unlikely to happen and even if it does, the code would
just mostly fail safely, so there was no attempt either to detect it or
prevent it by the code, which is something that might change in the future,
based on expected user feedback.
Reported-by: Guy Maurel <guy.j@maurel.de>
Helped-by: SZEDER Gábor <szeder.dev@gmail.com>
Helped-by: Randall Becker <rsbecker@nexbridge.com>
Helped-by: Phillip Wood <phillip.wood123@gmail.com>
Suggested-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Originally reported after release of v2.35.2 (and other maint branches)
for CVE-2022-24765 and blocking otherwise harmless commands that were
done using sudo in a repository that was owned by the user.
Add a new test script with very basic support to allow running git
commands through sudo, so a reproduction could be implemented and that
uses only `git status` as a proxy of the issue reported.
Note that because of the way sudo interacts with the system, a much
more complete integration with the test framework will require a lot
more work and that was therefore intentionally punted for now.
The current implementation requires the execution of a special cleanup
function which should always be kept as the last "test" or otherwise
the standard cleanup functions will fail because they can't remove
the root owned directories that are used. This also means that if
failures are found while running, the specifics of the failure might
not be kept for further debugging and if the test was interrupted, it
will be necessary to clean the working directory manually before
restarting by running:
$ sudo rm -rf trash\ directory.t0034-root-safe-directory/
The test file also uses at least one initial "setup" test that creates
a parallel execution directory under the "root" sub directory, which
should be used as top level directory for all repositories that are
used in this test file. Unlike all other tests the repository provided
by the test framework should go unused.
Special care should be taken when invoking commands through sudo, since
the environment is otherwise independent from what the test framework
setup and might have changed the values for HOME, SHELL and dropped
several relevant environment variables for your test. Indeed `git status`
was used as a proxy because it doesn't even require commits in the
repository to work and usually doesn't require much from the environment
to run, but a future patch will add calls to `git init` and that will
fail to honor the default branch name, unless that setting is NOT
provided through an environment variable (which means even a CI run
could fail that test if enabled incorrectly).
A new SUDO prerequisite is provided that does some sanity checking
to make sure the sudo command that will be used allows for passwordless
execution as root without restrictions and doesn't mess with git's
execution path. This matches what is provided by the macOS agents that
are used as part of GitHub actions and probably nowhere else.
Most of those characteristics make this test mostly only suitable for
CI, but it might be executed locally if special care is taken to provide
for all of them in the local configuration and maybe making use of the
sudo credential cache by first invoking sudo, entering your password if
needed, and then invoking the test with:
$ GIT_TEST_ALLOW_SUDO=YES ./t0034-root-safe-directory.sh
If it fails to run, then it means your local setup wouldn't work for the
test because of the configuration sudo has or other system settings, and
things that might help are to comment out sudo's secure_path config, and
make sure that the account you are using has no restrictions on the
commands it can run through sudo, just like is provided for the user in
the CI.
For example (assuming a username of marta for you) something probably
similar to the following entry in your /etc/sudoers (or equivalent) file:
marta ALL=(ALL:ALL) NOPASSWD: ALL
Reported-by: SZEDER Gábor <szeder.dev@gmail.com>
Helped-by: Phillip Wood <phillip.wood123@gmail.com>
Signed-off-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
With the addition of the safe.directory in 8959555ce
(setup_git_directory(): add an owner check for the top-level directory,
2022-03-02) released in v2.35.2, we are receiving feedback from a
variety of users about the feature.
Some users have a very large list of shared repositories and find it
cumbersome to add this config for every one of them.
In a more difficult case, certain workflows involve running Git commands
within containers. The container boundary prevents any global or system
config from communicating `safe.directory` values from the host into the
container. Further, the container almost always runs as a different user
than the owner of the directory in the host.
To simplify the reactions necessary for these users, extend the
definition of the safe.directory config value to include a possible '*'
value. This value implies that all directories are safe, providing a
single setting to opt-out of this protection.
Note that an empty assignment of safe.directory clears all previous
values, and this is already the case with the "if (!value || !*value)"
condition.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It seems that nothing is ever checking to make sure the safe directories
in the configs actually have the key safe.directory, so some unrelated
config that has a value with a certain directory would also make it a
safe directory.
Signed-off-by: Matheus Valadares <me@m28.io>
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It is difficult to change the ownership on a directory in our test
suite, so insert a new GIT_TEST_ASSUME_DIFFERENT_OWNER environment
variable to trick Git into thinking we are in a differently-owned
directory. This allows us to test that the config is parsed correctly.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When determining the length of the longest ancestor of a given path with
respect to to e.g. `GIT_CEILING_DIRECTORIES`, we special-case the root
directory by returning 0 (i.e. we pretend that the path `/` does not end
in a slash by virtually stripping it).
That is the correct behavior because when normalizing paths, the root
directory is special: all other directory paths have their trailing
slash stripped, but not the root directory's path (because it would
become the empty string, which is not a legal path).
However, this special-casing of the root directory in
`longest_ancestor_length()` completely forgets about Windows-style root
directories, e.g. `C:\`. These _also_ get normalized with a trailing
slash (because `C:` would actually refer to the current directory on
that drive, not necessarily to its root directory).
In fc56c7b34b (mingw: accomodate t0060-path-utils for MSYS2,
2016-01-27), we almost got it right. We noticed that
`longest_ancestor_length()` expects a slash _after_ the matched prefix,
and if the prefix already ends in a slash, the normalized path won't
ever match and -1 is returned.
But then that commit went astray: The correct fix is not to adjust the
_tests_ to expect an incorrect -1 when that function is fed a prefix
that ends in a slash, but instead to treat such a prefix as if the
trailing slash had been removed.
Likewise, that function needs to handle the case where it is fed a path
that ends in a slash (not only a prefix that ends in a slash): if it
matches the prefix (plus trailing slash), we still need to verify that
the path does not end there, otherwise the prefix is not actually an
ancestor of the path but identical to it (and we need to return -1 in
that case).
With these two adjustments, we no longer need to play games in t0060
where we only add `$rootoff` if the passed prefix is different from the
MSYS2 pseudo root, instead we also add it for the MSYS2 pseudo root
itself. We do have to be careful to skip that logic entirely for Windows
paths, though, because they do are not subject to that MSYS2 pseudo root
treatment.
This patch fixes the scenario where a user has set
`GIT_CEILING_DIRECTORIES=C:\`, which would be ignored otherwise.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
It poses a security risk to search for a git directory outside of the
directories owned by the current user.
For example, it is common e.g. in computer pools of educational
institutes to have a "scratch" space: a mounted disk with plenty of
space that is regularly swiped where any authenticated user can create
a directory to do their work. Merely navigating to such a space with a
Git-enabled `PS1` when there is a maliciously-crafted `/scratch/.git/`
can lead to a compromised account.
The same holds true in multi-user setups running Windows, as `C:\` is
writable to every authenticated user by default.
To plug this vulnerability, we stop Git from accepting top-level
directories owned by someone other than the current user. We avoid
looking at the ownership of each and every directories between the
current and the top-level one (if there are any between) to avoid
introducing a performance bottleneck.
This new default behavior is obviously incompatible with the concept of
shared repositories, where we expect the top-level directory to be owned
by only one of its legitimate users. To re-enable that use case, we add
support for adding exceptions from the new default behavior via the
config setting `safe.directory`.
The `safe.directory` config setting is only respected in the system and
global configs, not from repository configs or via the command-line, and
can have multiple values to allow for multiple shared repositories.
We are particularly careful to provide a helpful message to any user
trying to use a shared repository.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This function will be used in the next commit to prevent
`setup_git_directory()` from discovering a repository in a directory
that is owned by someone other than the current user.
Note: We cannot simply use `st.st_uid` on Windows just like we do on
Linux and other Unix-like platforms: according to
https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/stat-functions
this field is always zero on Windows (because Windows' idea of a user ID
does not fit into a single numerical value). Therefore, we have to do
something a little involved to replicate the same functionality there.
Also note: On Windows, a user's home directory is not actually owned by
said user, but by the administrator. For all practical purposes, it is
under the user's control, though, therefore we pretend that it is owned
by the user.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Build fix on Windows.
* cb/mingw-gmtime-r:
mingw: avoid fallback for {local,gm}time_r()
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
mingw-w64's pthread_unistd.h had a bug that mistakenly (because there is
no support for the *lockfile() functions required[1]) defined
_POSIX_THREAD_SAFE_FUNCTIONS and that was being worked around since
3ecd153a3b (compat/mingw: support MSys2-based MinGW build, 2016-01-14).
The bug was fixed in winphtreads, but as a side effect, leaves the
reentrant functions from time.h no longer visible and therefore breaks
the build.
Since the intention all along was to avoid using the fallback functions,
formalize the use of POSIX by setting the corresponding feature flag and
compile out the implementation for the fallback functions.
[1] https://unix.org/whitepapers/reentrant.html
Signed-off-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Acked-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint-2.22:
Git 2.22.5
Git 2.21.4
Git 2.20.5
Git 2.19.6
Git 2.18.5
Git 2.17.6
unpack_trees(): start with a fresh lstat cache
run-command: invalidate lstat cache after a command finished
checkout: fix bug that makes checkout follow symlinks in leading path
* maint-2.21:
Git 2.21.4
Git 2.20.5
Git 2.19.6
Git 2.18.5
Git 2.17.6
unpack_trees(): start with a fresh lstat cache
run-command: invalidate lstat cache after a command finished
checkout: fix bug that makes checkout follow symlinks in leading path
* maint-2.20:
Git 2.20.5
Git 2.19.6
Git 2.18.5
Git 2.17.6
unpack_trees(): start with a fresh lstat cache
run-command: invalidate lstat cache after a command finished
checkout: fix bug that makes checkout follow symlinks in leading path
* maint-2.19:
Git 2.19.6
Git 2.18.5
Git 2.17.6
unpack_trees(): start with a fresh lstat cache
run-command: invalidate lstat cache after a command finished
checkout: fix bug that makes checkout follow symlinks in leading path
* maint-2.18:
Git 2.18.5
Git 2.17.6
unpack_trees(): start with a fresh lstat cache
run-command: invalidate lstat cache after a command finished
checkout: fix bug that makes checkout follow symlinks in leading path
* maint-2.17:
Git 2.17.6
unpack_trees(): start with a fresh lstat cache
run-command: invalidate lstat cache after a command finished
checkout: fix bug that makes checkout follow symlinks in leading path
We really want to avoid relying on stale information.
Signed-off-by: Matheus Tavares <matheus.bernardino@usp.br>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
In the previous commit, we intercepted calls to `rmdir()` to invalidate
the lstat cache in the successful case, so that the lstat cache could
not have the idea that a directory exists where there is none.
The same situation can arise, of course, when a separate process is
spawned (most notably, this is the case in `submodule_move_head()`).
Obviously, we cannot know whether a directory was removed in that
process, therefore we must invalidate the lstat cache afterwards.
Note: in contrast to `lstat_cache_aware_rmdir()`, we invalidate the
lstat cache even in case of an error: the process might have removed a
directory and still have failed afterwards.
Co-authored-by: Matheus Tavares <matheus.bernardino@usp.br>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Before checking out a file, we have to confirm that all of its leading
components are real existing directories. And to reduce the number of
lstat() calls in this process, we cache the last leading path known to
contain only directories. However, when a path collision occurs (e.g.
when checking out case-sensitive files in case-insensitive file
systems), a cached path might have its file type changed on disk,
leaving the cache on an invalid state. Normally, this doesn't bring
any bad consequences as we usually check out files in index order, and
therefore, by the time the cached path becomes outdated, we no longer
need it anyway (because all files in that directory would have already
been written).
But, there are some users of the checkout machinery that do not always
follow the index order. In particular: checkout-index writes the paths
in the same order that they appear on the CLI (or stdin); and the
delayed checkout feature -- used when a long-running filter process
replies with "status=delayed" -- postpones the checkout of some entries,
thus modifying the checkout order.
When we have to check out an out-of-order entry and the lstat() cache is
invalid (due to a previous path collision), checkout_entry() may end up
using the invalid data and thrusting that the leading components are
real directories when, in reality, they are not. In the best case
scenario, where the directory was replaced by a regular file, the user
will get an error: "fatal: unable to create file 'foo/bar': Not a
directory". But if the directory was replaced by a symlink, checkout
could actually end up following the symlink and writing the file at a
wrong place, even outside the repository. Since delayed checkout is
affected by this bug, it could be used by an attacker to write
arbitrary files during the clone of a maliciously crafted repository.
Some candidate solutions considered were to disable the lstat() cache
during unordered checkouts or sort the entries before passing them to
the checkout machinery. But both ideas include some performance penalty
and they don't future-proof the code against new unordered use cases.
Instead, we now manually reset the lstat cache whenever we successfully
remove a directory. Note: We are not even checking whether the directory
was the same as the lstat cache points to because we might face a
scenario where the paths refer to the same location but differ due to
case folding, precomposed UTF-8 issues, or the presence of `..`
components in the path. Two regression tests, with case-collisions and
utf8-collisions, are also added for both checkout-index and delayed
checkout.
Note: to make the previously mentioned clone attack unfeasible, it would
be sufficient to reset the lstat cache only after the remove_subtree()
call inside checkout_entry(). This is the place where we would remove a
directory whose path collides with the path of another entry that we are
currently trying to check out (possibly a symlink). However, in the
interest of a thorough fix that does not leave Git open to
similar-but-not-identical attack vectors, we decided to intercept
all `rmdir()` calls in one fell swoop.
This addresses CVE-2021-21300.
Co-authored-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Matheus Tavares <matheus.bernardino@usp.br>
GitHub Actions is transitioning workflow steps that run on
'ubuntu-latest' from 18.04 to 20.04 [1].
This works fine in all steps except the static-analysis one, since
Coccinelle isn't available on Ubuntu focal (it is only available in the
universe suite).
Until Coccinelle can be installed from 20.04's main suite, pin the
static-analysis build to run on 18.04, where it can be installed by
default.
[1]: https://github.com/actions/virtual-environments/issues/1816
Reported-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our setting of GitHub CI test jobs were a bit too eager to give up
once there is even one failure found. Tweak the knob to allow
other jobs keep running even when we see a failure, so that we can
find more failures in a single run.
* pb/ci-matrix-wo-shortcut:
ci: do not cancel all jobs of a matrix if one fails
