conf: do exponential back off for sharing violation in hotfolder

Windows gives us notifications about writes to files in a directory, but
it does not give us notifications on when file handles are closed and
when we can expect to be able to grab a handle to it; this would be
racey at best. So, there always exists a race between the writer's last
call to WriteFile() and its eventual CloseHandle(). Work around this by
implementing a basic exponential back off of retrying the open call.

While we're at it, clean up the "file already exists" logic to remove a
basic toctou situation, and switch to using random temp file names in
order to handle better the case of saving a new file from two different
administrators at once.

Reported-by: Jim Salter <jim@jrs-s.net>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>

1 files changed, 1 insertions, 1 deletions

diff --git a/manager/service.go b/manager/service.go
index b671f79e..5ae8c9df 100644
--- a/manager/service.go
+++ b/manager/service.go
@@ -83,7 +83,7 @@ func (service *managerService) Execute(args []string, r <-chan svc.ChangeRequest
 		return
 	}
 
-	conf.RegisterStoreChangeCallback(func() { conf.MigrateUnencryptedConfigs() }) // Ignore return value for now, but could be useful later.
+	conf.RegisterStoreChangeCallback(func() { conf.MigrateUnencryptedConfigs(3) })
 	conf.RegisterStoreChangeCallback(IPCServerNotifyTunnelsChange)
 
 	procs := make(map[uint32]*os.Process)