The usual process flow is the main process opens and holds the lock to
the index, does its thing, perhaps spawning children during the course,
and then writes the resulting index out by releaseing the lock.
However, the lockfile interface uses atexit(3) to clean it up, without
regard to who actually created the lock. This typically leads to a
confusing behaviour of lock being released too early when the child
exits, and then the parent process when it calls commit_lockfile()
finds that it cannot unlock it.
This fixes the problem by recording who created and holds the lock, and
upon atexit(3) handler, child simply ignores the lockfile the parent
created.
Signed-off-by: Junio C Hamano <junkio@xxxxxxx>
---
cache.h | 1 +
lockfile.c | 6 +++++-
2 files changed, 6 insertions(+), 1 deletions(-)
diff --git a/cache.h b/cache.h
index 8c804cb..faddaf6 100644
--- a/cache.h
+++ b/cache.h
@@ -209,6 +209,7 @@ extern int refresh_cache(unsigned int flags);
struct lock_file {
struct lock_file *next;
+ pid_t owner;
char on_list;
char filename[PATH_MAX];
};
diff --git a/lockfile.c b/lockfile.c
index bed6b21..23db35a 100644
--- a/lockfile.c
+++ b/lockfile.c
@@ -8,8 +8,11 @@ static const char *alternate_index_output;
static void remove_lock_file(void)
{
+ pid_t me = getpid();
+
while (lock_file_list) {
- if (lock_file_list->filename[0])
+ if (lock_file_list->owner == me &&
+ lock_file_list->filename[0])
unlink(lock_file_list->filename);
lock_file_list = lock_file_list->next;
}
@@ -28,6 +31,7 @@ static int lock_file(struct lock_file *lk, const char *path)
sprintf(lk->filename, "%s.lock", path);
fd = open(lk->filename, O_RDWR | O_CREAT | O_EXCL, 0666);
if (0 <= fd) {
+ lk->owner = getpid();
if (!lk->on_list) {
lk->next = lock_file_list;
lock_file_list = lk;
--
1.5.1.1.948.g9f6f
-
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html