Hi, Michael-- Thanks for the quickly reply.
If the metadata on the cache device was NOT correctly read, cached_dev_read_error() will not be called, so the recovery check will not be executed. so this patch is safe. s->iop.error was not set when faild to read metadata on the cache device, so the cached_dev_bio_complete() will be called instead of cached_dev_read_error() 2017-11-17 12:02 GMT+08:00 Michael Lyle <[email protected]>: > Hi, Rui Hua-- > > On 11/16/2017 07:51 PM, Rui Hua wrote: >> In this patch, we use s->read_dirty_data to judge whether the read >> request hit dirty data in cache device, it is safe to reread data from >> the backing device when the read request hit clean data. This can not >> only handle cache read race, but also recover data when failed read >> request from cache device. > > I haven't read over all of this yet, to understand the read race. But > this change can't be safe-- read_dirty_data only is set to true if the > metadata on the cache device is correctly read. If that fails, the flag > may not be set and we could read stale data on the backing device. > > Mike
