Some platforms (e.g. Windows) provide API functions to resolve paths
much quicker. Let's offer a way to short-cut `strbuf_realpath()` on
those platforms.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Always use the internal "use_weak" random seed when initializing
the "mimalloc" heap when statically linked on Windows.
The imported "mimalloc" routines support several random sources
to seed the heap data structures, including BCrypt.dll and
RtlGenRandom. Crashes have been reported when using BCrypt.dll
if it initialized during an `atexit()` handler function. Granted,
such DLL initialization should not happen in an atexit handler,
but yet the crashes remain.
It should be noted that on Windows when statically linked, the
mimalloc startup code (called by the GCC CRT to initialize static
data prior to calling `main()`) always uses the internal "weak"
random seed. "mimalloc" does not try to load an alternate
random source until after the OS initialization has completed.
Heap data is stored in `__declspec(thread)` TLS data and in theory
each Git thread will have its own heap data. However, testing
shows that the "mimalloc" library doesn't actually call
`os_random_buf()` (to load a new random source) when creating these
new per-thread heap structures.
However, if an atexit handler is forced to run on a non-main
thread, the "mimalloc" library *WILL* try to create a new heap
and seed it with `os_random_buf()`. (The reason for this is still
a mystery to this author.) The `os_random_buf()` call can cause
the (previously uninitialized BCrypt.dll library) to be dynamically
loaded and a call made into it. Crashes have been reported in
v2.40.1.vfs.0.0 while in this call.
As a workaround, the fix here forces the use of the internal
"use_weak" random code for the subsequent `os_random_buf()` calls.
Since we have been using that random generator for the majority
of the program, it seems safe to use it for the final few mallocs
in the atexit handler (of which there really shouldn't be that many.
Signed-off-by: Jeff Hostetler <jeffhostetler@github.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
NTFS junctions are somewhat similar in spirit to Unix bind mounts: they
point to a different directory and are resolved by the filesystem
driver. As such, they appear to `lstat()` as if they are directories,
not as if they are symbolic links.
_Any_ user can create junctions, while symbolic links can only be
created by non-administrators in Developer Mode on Windows 10. Hence
NTFS junctions are much more common "in the wild" than NTFS symbolic
links.
It was reported in https://github.com/git-for-windows/git/issues/2481
that adding files via an absolute path that traverses an NTFS junction:
since 1e64d18 (mingw: do resolve symlinks in `getcwd()`), we resolve not
only symbolic links but also NTFS junctions when determining the
absolute path of the current directory. The same is not true for `git
add <file>`, where symbolic links are resolved in `<file>`, but not NTFS
junctions.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
By defining `USE_MIMALLOC`, Git can now be compiled with that
nicely-fast and small allocator.
Note that we have to disable a couple `DEVELOPER` options to build
mimalloc's source code, as it makes heavy use of declarations after
statements, among other things that disagree with Git's conventions.
We even have to silence some GCC warnings in non-DEVELOPER mode. For
example, the `-Wno-array-bounds` flag is needed because in `-O2` builds,
trying to call `NtCurrentTeb()` (which `_mi_thread_id()` does on
Windows) causes the bogus warning about a system header, likely related
to https://sourceforge.net/p/mingw-w64/mailman/message/37674519/ and to
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99578:
C:/git-sdk-64-minimal/mingw64/include/psdk_inc/intrin-impl.h:838:1:
error: array subscript 0 is outside array bounds of 'long long unsigned int[0]' [-Werror=array-bounds]
838 | __buildreadseg(__readgsqword, unsigned __int64, "gs", "q")
| ^~~~~~~~~~~~~~
Also: The `mimalloc` library uses C11-style atomics, therefore we must
require that standard when compiling with GCC if we want to use
`mimalloc` (instead of requiring "only" C99). This is what we do in the
CMake definition already, therefore this commit does not need to touch
`contrib/buildsystems/`.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
We want to compile mimalloc's source code as part of Git, rather than
requiring the code to be built as an external library: mimalloc uses a
CMake-based build, which is not necessarily easy to integrate into the
flavors of Git for Windows (which will be the main benefitting port).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This commit imports mimalloc's source code as per v2.0.9, fetched from
the tag at https://github.com/microsoft/mimalloc.
The .c files are from the src/ subdirectory, and the .h files from the
include/ subdirectory. We will subsequently modify the source code to
accommodate building within Git's context.
Since we plan on using the `mi_*()` family of functions, we skip the
C++-specific source code, some POSIX compliant functions to interact
with mimalloc, and the code that wants to support auto-magic overriding
of the `malloc()` function (mimalloc-new-delete.h, alloc-posix.c,
mimalloc-override.h, alloc-override.c, alloc-override-osx.c,
alloc-override-win.c and static.c).
To appease the `check-whitespace` job of Git's Continuous Integration,
this commit was washed one time via `git rebase --whitespace=fix`.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
We are about to vendor in `mimalloc`'s source code which we will want to
include `git-compat-util.h` after defining that constant.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This topic branch extends the protections introduced for Git GUI's
CVE-2022-41953 to cover `gitk`, too.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
While Git for Windows does not _ship_ Python (in order to save on
bandwidth), MSYS2 provides very fine Python interpreters that users can
easily take advantage of, by using Git for Windows within its SDK.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Windows' equivalent to "bind mounts", NTFS junction points, can be
unlinked without affecting the mount target. This is clearly what users
expect to happen when they call `git clean -dfx` in a worktree that
contains NTFS junction points: the junction should be removed, and the
target directory of said junction should be left alone (unless it is
inside the worktree).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
When specifying an absolute path without a drive prefix, we convert that
path internally. Let's make sure that we handle that case properly, too
;-)
This fixes the command
git clone https://github.com/git-for-windows/git \G4W
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
After importing anything with fast-import, we should always let the
garbage collector do its job, since the objects are written to disk
inefficiently.
This brings down an initial import of http://selenic.com/hg from about
230 megabytes to about 14.
In the future, we may want to make this configurable on a per-remote
basis, or maybe teach fast-import about it in the first place.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
The mingw-w64 GCC seems to link implicitly to libwinpthread, which does
implement a pthread emulation (that is more complete than Git's). Let's
keep preferring Git's.
To avoid linker errors where it thinks that the `pthread_self` and the
`pthread_create` symbols are defined twice, let's give our version a
`win32_` prefix, just like we already do for `pthread_join()`.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This topic branch fixes a vulnerability in Git GUI's "clone" feature
(tracked as CVE-2022-41953) that was graded with a CVSS Score 8.6/10
(high).
These patches were backported to Git GUI in
https://github.com/prati0100/git-gui/pull/85
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Just like CVE-2022-41953 for Git GUI, there exists a vulnerability of
`gitk` where it looks for `taskkill.exe` in the current directory before
searching `PATH`.
Note that the many `exec git` calls are unaffected, due to an obscure
quirk in Tcl's `exec` function. Typically, `git.exe` lives next to
`wish.exe` (i.e. the program that is run to execute `gitk` or Git GUI)
in Git for Windows, and that is the saving grace for `git.exe because
`exec` searches the directory where `wish.exe` lives even before the
current directory, according to
https://www.tcl-lang.org/man/tcl/TclCmd/exec.htm#M24:
If a directory name was not specified as part of the application
name, the following directories are automatically searched in
order when attempting to locate the application:
The directory from which the Tcl executable was loaded.
The current directory.
The Windows 32-bit system directory.
The Windows home directory.
The directories listed in the path.
The same is not true, however, for `taskkill.exe`: it lives in the
Windows system directory (never mind the 32-bit, Tcl's documentation is
outdated on that point, it really means `C:\Windows\system32`).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
It seems to be not exactly rare on Windows to install NTFS junction
points (the equivalent of "bind mounts" on Linux/Unix) in worktrees,
e.g. to map some development tools into a subdirectory.
In such a scenario, it is pretty horrible if `git clean -dfx` traverses
into the mapped directory and starts to "clean up".
Let's just not do that. Let's make sure before we traverse into a
directory that it is not a mount point (or junction).
This addresses https://github.com/git-for-windows/git/issues/607
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
On Windows, there are several categories of absolute paths. One such
category starts with a backslash and is implicitly relative to the
drive associated with the current working directory. Example:
c:
git clone https://github.com/git-for-windows/git \G4W
should clone into C:\G4W.
There is currently a problem with that, in that mingw_mktemp() does not
expect the _wmktemp() function to prefix the absolute path with the
drive prefix, and as a consequence, the resulting path does not fit into
the originally-passed string buffer. The symptom is a "Result too large"
error.
Reported by Juan Carlos Arevalo Baeza.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
As per https://www.tcl.tk/man/tcl8.6/TclCmd/exec.html#M23, Tcl's `exec`
function goes out of its way to imitate the highly dangerous path lookup
of `cmd.exe`, but _of course_ only on Windows:
If a directory name was not specified as part of the application
name, the following directories are automatically searched in
order when attempting to locate the application:
The directory from which the Tcl executable was loaded.
The current directory.
The Windows 32-bit system directory.
The Windows home directory.
The directories listed in the path.
The dangerous part is the second item, of course: `exec` _prefers_
executables in the current directory to those that are actually in the
`PATH`.
It is almost as if people wanted to Windows users vulnerable,
specifically.
To avoid that, Git GUI already has the `_which` function that does not
imitate that dangerous practice when looking up executables in the
search path.
However, Git GUI currently fails to use that function e.g. when trying to
execute `aspell` for spell checking.
That is not only dangerous but combined with Tcl's unfortunate default
behavior and with the fact that Git GUI tries to spell-check a
repository just after cloning, leads to a critical Remote Code Execution
vulnerability.
Let's override both `exec` and `open` to always use `_which` instead of
letting Tcl perform the path lookup, to prevent this attack vector.
This addresses CVE-2022-41953.
For more details, see
https://github.com/git-for-windows/git/security/advisories/GHSA-v4px-mx59-w99c
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
We are about to make use of the `_which` function to address
CVE-2022-41953 by overriding Tcl/Tk's unsafe PATH lookup on Windows.
In preparation for that, let's move it close to the top of the file to
make sure that even early `exec` calls that happen during the start-up
of Git GUI benefit from the fix.
This commit is best viewed with `--color-moved`.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
This happens only when the corresponding commits are not exported in
the current fast-export run. This can happen either when the relevant
commit is already marked, or when the commit is explicitly marked
as UNINTERESTING with a negative ref by another argument.
This breaks fast-export basec remote helpers.
Signed-off-by: Sverre Rabbelier <srabbelier@gmail.com>
We need these in `_which` and they should be defined before that
function's definition.
This commit is best viewed with `--color-moved`.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
The `is_Cygwin` function is used, among other things, to determine
how executables are discovered in the `PATH` list by the `_which` function.
We are about to change the behavior of the `_which` function on Windows
(but not Cygwin): On Windows, we want it to ignore empty elements of the
`PATH` instead of treating them as referring to the current directory
(which is a "legacy feature" according to
https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html#tag_08_03,
but apparently not explicitly deprecated, the POSIX documentation is
quite unclear on that even if the Cygwin project itself considers it to
be deprecated: https://github.com/cygwin/cygwin/commit/fc74dbf22f5c).
This is important because on Windows, `exec` does something very unsafe
by default (unless we're running a Cygwin version of Tcl, which follows
Unix semantics).
However, we try to `exec` something _inside_ `is_Cygwin` to determine
whether we're running within Cygwin or not, i.e. before we determined
whether we need to handle `PATH` specially or not. That's a Catch-22.
Therefore, and because it is much cleaner anyway, use the
`$::tcl_platform(os)` value which is guaranteed to start with `CYGWIN_`
when running a Cygwin variant of Tcl/Tk, instead of executing `cygpath
--windir`.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
When looking up an executable via the `_which` function, Git GUI
imitates the `execlp()` strategy where the environment variable `PATH`
is interpreted as a list of paths in which to search.
For historical reasons, stemming from the olden times when it was
uncommon to download a lot of files from the internet into the current
directory, empty elements in this list are treated as if the current
directory had been specified.
Nowadays, of course, this treatment is highly dangerous as the current
directory often contains files that have just been downloaded and not
yet been inspected by the user. Unix/Linux users are essentially
expected to be very, very careful to simply not add empty `PATH`
elements, i.e. not to make use of that feature.
On Windows, however, it is quite common for `PATH` to contain empty
elements by mistake, e.g. as an unintended left-over entry when an
application was installed from the Windows Store and then uninstalled
manually.
While it would probably make most sense to safe-guard not only Windows
users, it seems to be common practice to ignore these empty `PATH`
elements _only_ on Windows, but not on other platforms.
Sadly, this practice is followed inconsistently between different
software projects, where projects with few, if any, Windows-based
contributors tend to be less consistent or even "blissful" about it.
Here is a non-exhaustive list:
Cygwin:
It specifically "eats" empty paths when converting path lists to
POSIX: https://github.com/cygwin/cygwin/commit/753702223c7d
I.e. it follows the common practice.
PowerShell:
It specifically ignores empty paths when searching the `PATH`.
The reason for this is apparently so self-evident that it is not
even mentioned here:
https://learn.microsoft.com/en-us/powershell/module/microsoft.powershell.core/about/about_environment_variables#path-information
I.e. it follows the common practice.
CMD:
Oh my, CMD. Let's just forget about it, nobody in their right
(security) mind takes CMD as inspiration. It is so unsafe by
default that we even planned on dropping `Git CMD` from Git for
Windows altogether, and only walked back on that plan when we
found a super ugly hack, just to keep Git's users secure by
default:
https://github.com/git-for-windows/MINGW-packages/commit/82172388bb51
So CMD chooses to hide behind the battle cry "Works as
Designed!" that all too often leaves users vulnerable. CMD is
probably the most prominent project whose lead you want to avoid
following in matters of security.
Win32 API (`CreateProcess()`)
Just like CMD, `CreateProcess()` adheres to the original design
of the path lookup in the name of backward compatibility (see
https://learn.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-createprocessw
for details):
If the file name does not contain a directory path, the
system searches for the executable file in the following
sequence:
1. The directory from which the application loaded.
2. The current directory for the parent process.
[...]
I.e. the Win32 API itself chooses backwards compatibility over
users' safety.
Git LFS:
There have been not one, not two, but three security advisories
about Git LFS executing executables from the current directory by
mistake. As part of one of them, a change was introduced to stop
treating empty `PATH` elements as equivalent to `.`:
https://github.com/git-lfs/git-lfs/commit/7cd7bb0a1f0d
I.e. it follows the common practice.
Go:
Go does not follow the common practice, and you can think about
that what you want:
https://github.com/golang/go/blob/go1.19.3/src/os/exec/lp_windows.go#L114-L135https://github.com/golang/go/blob/go1.19.3/src/path/filepath/path_windows.go#L108-L137
Git Credential Manager:
It tries to imitate Git LFS, but unfortunately misses the empty
`PATH` element handling. As of time of writing, this is in the
process of being fixed:
https://github.com/GitCredentialManager/git-credential-manager/pull/968
So now that we have established that it is a common practice to ignore
empty `PATH` elements on Windows, let's assess this commit's change
using Schneier's Five-Step Process
(https://www.schneier.com/crypto-gram/archives/2002/0415.html#1):
Step 1: What problem does it solve?
It prevents an entire class of Remote Code Execution exploits via
Git GUI's `Clone` functionality.
Step 2: How well does it solve that problem?
Very well. It prevents the attack vector of luring an unsuspecting
victim into cloning an executable into the worktree root directory
that Git GUI immediately executes.
Step 3: What other security problems does it cause?
Maybe non-security problems: If a project (ab-)uses the unsafe
`PATH` lookup. That would not only be unsafe, though, but
fragile in the first place because it would break when running
in a subdirectory. Therefore I would consider this a scenario
not worth keeping working.
Step 4: What are the costs of this measure?
Almost nil, except for the time writing up this commit message
;-)
Step 5: Given the answers to steps two through four, is the security
measure worth the costs?
Yes. Keeping Git's users Secure By Default is worth it. It's a
tiny price to pay compared to the damages even a single
successful exploit can cost.
So let's follow that common practice in Git GUI, too.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
* 'master' of github.com:git/git: (34 commits)
Git 2.42-rc2
t4053: avoid writing to unopened pipe
t4053: avoid race when killing background processes
Git 2.42-rc1
git maintenance: avoid console window in scheduled tasks on Windows
win32: add a helper to run `git.exe` without a foreground window
t9001: remove excessive GIT_SEND_EMAIL_NOTTY=1
mv: handle lstat() failure correctly
parse-options: disallow negating OPTION_SET_INT 0
repack: free geometry struct
send-email: avoid creating more than one Term::ReadLine object
send-email: drop FakeTerm hack
t0040: declare non-tab indentation to be okay in this script
advice: handle "rebase" in error_resolve_conflict()
A few more topics before -rc1
mailmap: change primary address for Glen Choo
gitignore: ignore clangd .cache directory
docs: update when `git bisect visualize` uses `gitk`
compat/mingw: implement a native locate_in_PATH()
run-command: conditionally define locate_in_PATH()
...
Windows updates.
* ds/maintenance-on-windows-fix:
git maintenance: avoid console window in scheduled tasks on Windows
win32: add a helper to run `git.exe` without a foreground window
Adjust to newer Term::ReadLine to prevent it from breaking
the interactive prompt code in send-email.
* jk/send-email-with-new-readline:
send-email: avoid creating more than one Term::ReadLine object
send-email: drop FakeTerm hack
Developer support to detect meaningless combination of options.
* rs/parse-opt-forbid-set-int-0-without-noneg:
parse-options: disallow negating OPTION_SET_INT 0
The continuations of the compound command were indented as if they were
continuations of the embedded pipe, which was misleading.
Signed-off-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This fixes an occasional hang I see when running t4053 with
--verbose-log using dash.
Commit 1e3f26542a (diff --no-index: support reading from named pipes,
2023-07-05) added a test that "diff --no-index" will complain when
comparing a named pipe and a directory. The minimum we need to test this
is to mkfifo the pipe, and then run "git diff --no-index pipe some_dir".
But the test does one thing more: it spawns a background shell process
that opens the pipe for writing, like this:
{
(>pipe) &
} &&
This extra writer _could_ be useful if Git misbehaves and tries to open
the pipe for reading. Without the writer, Git would block indefinitely
and the test would never end. But since we do not have such a bug, Git
does not open the pipe and it is the writing process which will block
indefinitely, since there are no readers. The test addresses this by
running "kill $!" in a test_when_finished block. Since the writer should
be blocking forever, this kill command will reliably find it waiting.
However, this seems to be somewhat racy, in that the writing process
sometimes hangs around even after the "kill". In a normal run of the
test script without options, this doesn't have any effect; the
main test script completes anyway. But with --verbose-log, we spawn a
"tee" process that reads the script output, and it won't end until all
descriptors pointing to its input pipe are closed. And the background
process that is hanging around still has its stderr, etc, pointed into
that pipe.
You can reproduce the situation like this:
cd t
./t4053-diff-no-index.sh --verbose-log --stress
Let that run for a few minutes, and then you'll find that some of the
runs have hung. For example, at 11:53, I ran:
$ ps xk start o pid,start,command | grep tee | head
713459 11:48:06 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-9.out
713527 11:48:06 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-15.out
719434 11:48:07 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-1.out
728117 11:48:08 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-5.out
738738 11:48:09 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-31.out
739457 11:48:09 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-27.out
744432 11:48:10 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-21.out
744471 11:48:10 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-29.out
761961 11:48:12 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-0.out
812299 11:48:19 tee -a /home/peff/compile/git/t/test-results/t4053-diff-no-index.stress-8.out
All of these have been hung for several minutes. We can investigate one
and see that it's waiting to get EOF on its input:
$ strace -p 713459
strace: Process 713459 attached
read(0,
^C
Who else has that descriptor open?
$ lsof -a -p 713459 -d 0 +E
COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME
tee 713459 peff 0r FIFO 0,13 0t0 3943636 pipe 719203,sh,5w 719203,sh,7w 719203,sh,12w 719203,sh,13w
sh 719203 peff 5w FIFO 0,13 0t0 3943636 pipe 713459,tee,0r 719203,sh,7w 719203,sh,12w 719203,sh,13w
sh 719203 peff 7w FIFO 0,13 0t0 3943636 pipe 713459,tee,0r 719203,sh,5w 719203,sh,12w 719203,sh,13w
sh 719203 peff 12w FIFO 0,13 0t0 3943636 pipe 713459,tee,0r 719203,sh,5w 719203,sh,7w 719203,sh,13w
sh 719203 peff 13w FIFO 0,13 0t0 3943636 pipe 713459,tee,0r 719203,sh,5w 719203,sh,7w 719203,sh,12w
It's a shell, presumably a subshell spawned by the main script. Though
it may seem odd, having the same descriptor open several times is not
unreasonable (they're all basically the original stdout/stderr of the
script that has been copied). And they should all close when the process
exits. So what's it doing? Curiously, it will exit as soon as we strace
it:
$ strace -s 64 -p 719203
strace: Process 719203 attached
openat(AT_FDCWD, "pipe", O_WRONLY|O_CREAT|O_TRUNC, 0666) = -1 ENOENT (No such file or directory)
write(2, "./t4053-diff-no-index.sh: 7: eval: ", 35) = 35
write(2, "cannot create pipe: Directory nonexistent", 41) = 41
write(2, "\n", 1) = 1
exit_group(2) = ?
+++ exited with 2 +++
I think what happens is this:
- it is blocking in the openat() call for the pipe, as we expect (so
this is definitely the backgrounded subshell mentioned above)
- strace sends signals (probably STOP/CONT); those cause the kernel to
stop blocking, but libc will restart the system call automatically
- by this time, the "pipe" fifo is gone, so we'll actually try to
create a regular file. But of course the surrounding directory is
gone, too! So we get ENOENT, and then exit as normal.
So the blocking is something we expect to happen. But what we didn't
expect is for the process to still exist at all! It should have been
killed earlier when the parent process called "kill", but it wasn't. And
we can't catch the race at this point, because it happened much earlier.
One can guess, though, that there is some race with the shell setting up
the signal handling in the backgrounded subshell, and possibly blocking
or ignoring signals at the time that the "kill" is received. Curiously,
the race does not seem to happen if I use "bash" instead of "dash", so
presumably bash's setup here is more atomic.
One fix might be to try killing the subshell more aggressively, either
using SIGKILL, or looping on kill/wait. But that seems complex and
likely to introduce new problems/races. Instead, we can observe that the
writer is not needed at all. Git will notice the pipe via stat() before
it is ever opened. So we can simply drop the writer subshell entirely.
If we ever changed Git to open the path and fstat() it, this would
result in the test hanging. But we're not likely to do that. After all,
we have to stat() paths to see if they are openable at all (e.g., it
could be a directory), so this seems like a low risk. And anybody who
does make such a change will immediately see the issue, as Git would
hang consistently.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If both directories D1 and D2 already exists, and further there is a
filesystem entity D2/D1, "git mv D1 D2" would fail, and we get an
error message that says:
"cannot move directory over file, source=D1, destination=D2/D1"
regardless of the type of existing "D2/D1". If it is a file, the
message is correct, but if it is a directory, it is not (we could
make the D2/D1 directory a union of its original contents and what
was in D1/, but that is not what we do).
The code that decies to issue the error message only checks for
existence of "D2/D1" and does not care what kind of thing sits at
the path.
Rephrase the message to say
"destination already exists, source=D1, destination=D2/D1"
that would be suitable for any kind of thing being in the way.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Set the requires-full-index to false for "check-attr".
Add a test to ensure that the index is not expanded whether the files
are outside or inside the sparse-checkout cone when the sparse index is
enabled.
The `p2000` tests demonstrate a ~63% execution time reduction for
'git check-attr' using a sparse index.
Test before after
-----------------------------------------------------------------------
2000.106: git check-attr -a f2/f4/a (full-v3) 0.05 0.05 +0.0%
2000.107: git check-attr -a f2/f4/a (full-v4) 0.05 0.05 +0.0%
2000.108: git check-attr -a f2/f4/a (sparse-v3) 0.04 0.02 -50.0%
2000.109: git check-attr -a f2/f4/a (sparse-v4) 0.04 0.01 -75.0%
Helped-by: Victoria Dye <vdye@github.com>
Signed-off-by: Shuqi Liang <cheskaqiqi@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before this patch, git check-attr was unable to read the attributes from
a .gitattributes file within a sparse directory. The original comment
was operating under the assumption that users are only interested in
files or directories inside the cones. Therefore, in the original code,
in the case of a cone-mode sparse-checkout, we didn't load the
.gitattributes file.
However, this behavior can lead to missing attributes for files inside
sparse directories, causing inconsistencies in file handling.
To resolve this, revise 'git check-attr' to allow attribute reading for
files in sparse directories from the corresponding .gitattributes files:
1.Utilize path_in_cone_mode_sparse_checkout() and index_name_pos_sparse
to check if a path falls within a sparse directory.
2.If path is inside a sparse directory, employ the value of
index_name_pos_sparse() to find the sparse directory containing path and
path relative to sparse directory. Proceed to read attributes from the
tree OID of the sparse directory using read_attr_from_blob().
3.If path is not inside a sparse directory，ensure that attributes are
fetched from the index blob with read_blob_data_from_index().
Change the test 'check-attr with pathspec outside sparse definition' to
'test_expect_success' to reflect that the attributes inside a sparse
directory can now be read. Ensure that the sparse index case works
correctly for git check-attr to illustrate the successful handling of
attributes within sparse directories.
Helped-by: Victoria Dye <vdye@github.com>
Signed-off-by: Shuqi Liang <cheskaqiqi@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add tests for `git check-attr`, make sure attribute file does get read
from index when path is either inside or outside of sparse-checkout
definition.
Add a test named 'diff --check with pathspec outside sparse definition'.
It starts by disabling the trailing whitespace and space-before-tab
checks using the core. whitespace configuration option. Then, it
specifically re-enables the trailing whitespace check for a file located
in a sparse directory by adding a whitespace=trailing-space rule to the
.gitattributes file within that directory. Next, create and populate the
folder1 directory, and then add the .gitattributes file to the index.
Edit the contents of folder1/a, add it to the index, and proceed to
"re-sparsify" 'folder1/' with 'git sparse-checkout reapply'. Finally,
use 'git diff --check --cached' to compare the 'index vs. HEAD',
ensuring the correct application of the attribute rules even when the
file's path is outside the sparse-checkout definition.
Mark the two tests 'check-attr with pathspec outside sparse definition'
and 'diff --check with pathspec outside sparse definition' as
'test_expect_failure' to reflect an existing issue where the attributes
inside a sparse directory are ignored. Ensure that the 'check-attr'
command fails to read the required attributes to demonstrate this
expected failure.
Helped-by: Victoria Dye <vdye@github.com>
Signed-off-by: Shuqi Liang <cheskaqiqi@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When running 'git maintenance start', the current pattern is to
configure global config settings to enable maintenance on the current
repository and set 'maintenance.auto' to false and _then_ to set up the
schedule with the system scheduler.
This has a problematic error condition: if the scheduler fails to
initialize, the repository still will not use automatic maintenance due
to the 'maintenance.auto' setting.
Fix this gap by swapping the order of operations. If Git fails to
initialize maintenance, then the config changes should never happen.
Reported-by: Phillip Wood <phillip.wood123@gmail.com>
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The 'git maintenance run' command prevents concurrent runs in the same
repository using a 'maintenance.lock' file. However, when using systemd
the hourly maintenance runs the same time as the daily and weekly runs.
(Similarly, daily maintenance runs at the same time as weekly
maintenance.) These competing commands result in some maintenance not
actually being run.
This overlap was something we could not fix until we made the recent
change to not use the builting 'hourly', 'daily', and 'weekly' schedules
in systemd. We can adjust the schedules such that:
1. Hourly runs avoid the 0th hour.
2. Daily runs avoid Monday.
This will keep maintenance runs from colliding when using systemd.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The get_random_minute() method was created to allow maintenance
schedules to be fixed to a random minute of the hour. This randomness is
only intended to spread out the load from a number of clients, but each
client should have an hour between each maintenance cycle.
Add this random minute to the systemd integration.
This integration is more complicated than similar changes for other
schedulers because of a neat trick that systemd allows: templating.
The previous implementation generated two template files with names
of the form 'git-maintenance@.(timer|service)'. The '.timer' or
'.service' indicates that this is a template that is picked up when we
later specify '...@<schedule>.timer' or '...@<schedule>.service'. The
'<schedule>' string is then used to insert into the template both the
'OnCalendar' schedule setting and the '--schedule' parameter of the
'git maintenance run' command.
In order to set these schedules to a given minute, we can no longer use
the 'hourly', 'daily', or 'weekly' strings for '<schedule>' and instead
need to abandon the template model for the .timer files. We can still
use templates for the .service files. For this reason, we split these
writes into two methods.
Modify the template with a custom schedule in the 'OnCalendar' setting.
This schedule has some interesting differences from cron-like patterns,
but is relatively easy to figure out from context. The one that might be
confusing is that '*-*-*' is a date-based pattern, but this must be
omitted when using 'Mon' to signal that we care about the day of the
week. Monday is used since that matches the day used for the 'weekly'
schedule used previously.
Now that the timer files are not templates, we might want to abandon the
'@' symbol in the file names. However, this would cause users with
existing schedules to get two competing schedules due to different
names. The work to remove the old schedule name is one thing that we can
avoid by keeping the '@' symbol in our unit names. Since we are locked
into this name, it makes sense that we keep the template model for the
.service files.
The rest of the change involves making sure we are writing these .timer
and .service files before initializing the schedule with 'systemctl' and
deleting the files when we are done. Some changes are also made to share
the random minute along with a single computation of the execution path
of the current Git executable.
In addition, older Git versions may have written a
'git-maintenance@.timer' template file. Be sure to remove this when
successfully enabling maintenance (or disabling maintenance).
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The systemd_timer_write_unit_templates() method writes a single template
that is then used to start the hourly, daily, and weekly schedules with
systemd.
However, in order to schedule systemd maintenance on a given minute,
these templates need to be replaced with specific schedules for each of
these jobs.
Before modifying the schedules, move the writing method above the
systemd_timer_enable_unit() method, so we can write a specific schedule
for each unit.
The diff is computed smaller by showing systemd_timer_enable_unit() and
systemd_timer_delete_units() move instead of
systemd_timer_write_unit_templates() and
systemd_timer_delete_unit_templates().
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The get_random_minute() method was created to allow maintenance
schedules to be fixed to a random minute of the hour. This randomness is
only intended to spread out the load from a number of clients, but each
client should have an hour between each maintenance cycle.
Add this random minute to the cron integration.
The cron schedule specification starts with a minute indicator, which
was previously inserted as the "0" string but now takes the given minute
as an integer parameter.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
