Hi list. It's been a while.
I can reliably reproduce a segfault in git grep when searching the
object DB. It is caused by a race when threads > 2.
I have a patch that fixes the problem, but I don't know exactly why.
Can someone else explain it and/or offer a better solution?
====
diff --git packfile.c packfile.c
index e69012e7f2..52b7b54aeb 100644
--- packfile.c
+++ packfile.c
@@ -1812,7 +1812,9 @@ void *unpack_entry(struct repository *r, struct
packed_git *p, off_t obj_offset,
if (!base)
continue;
+ obj_read_lock();
delta_data = unpack_compressed_entry(p, &w_curs,
curpos, delta_size);
+ obj_read_unlock();
if (!delta_data) {
error("failed to unpack compressed delta "
====
My notes thus far:
I have seen it fail with `--threads 3`, but it fails very reliably for
me with `--threads 20`.
My repo is large (7GB) which gives grep enough to do. I'm not able to
make it fail on git.git; it seems like it's too small to run into this
problem.
The problem bisects to 1d1729caeb. It was somewhat hard to find this,
though, because git grep disables --threads when reading from the
object store until another patch later in the same series.
====
$ git bisect run sh -c \
"sed -i -e 's/list.nr || cached || show_in_pager/list.nr ||
show_in_pager/' builtin/grep.c || exit 129;
make -j 30 || exit 125;
git checkout HEAD -- builtin/grep.c ;
./bin-wrappers/git -C ~/git/purity grep --threads 20 --cached
'target-regex' || exit 1"
1d1729caebd41b340dd8dd61057f613da4df526c is the first bad commit
commit 1d1729caebd41b340dd8dd61057f613da4df526c
Author: Matheus Tavares <matheus.bernardino@xxxxxx>
Date: Wed Jan 15 23:39:54 2020 -0300
grep: replace grep_read_mutex by internal obj read lock
Here's what I'm testing with:
$ ./bin-wrappers/git --version
git version 2.28.0.585.ge1cfff6765
I have found that the crash always happens in cache_or_unpack_entry.
The thread that segfaults is always in patch_delta, and there are
always at least two threads inside unpack_compressed_entry at the time
of the failure.
I was able to get it to crash with just three threads one time. I
copied a log with the stack trace of all (4) threads during that run
to a gist [1].
Another log showing four different segfaults when run with 40 threads
is also in that gist [1].
[1] https://gist.github.com/phord/02e84d003688baa493b978110225d443
