Mikulas, > I/Os shouldn't be returned with -ENOMEM. If they are, you can treat it as > a hard error. It seems to be blkdev_issue_discard returns -ENOMEM when bio_alloc fails, for example. Waiting for a second and we can alloc the memory is my idea for handling -ENOMEM returned. > Blocking I/O until the admin turns a specific variable isn't too > reliable. > > Think of this case - your driver detects I/O error and blocks all I/Os. > The admin tries to log in. The login process needs memory. To fulfill this > memory need, the login process writes out some dirty pages. Those writes > are blocked by your driver - in the result, the admin is not able to log > in and flip the switch to unblock I/Os. > > Blocking I/O indefinitely isn't good because any system activity > (including typing commands into shell) may wait on this I/O. I understand the problem. But, what should I do then? Since writeboost is a cache software, it loses consistency if we ignore the cache at all in its returning I/O error. Go panic in that case is also inappropriate (But, inaccessibility to the storage will eventually halt the whole system. If so, go panic might be an acceptable solution). I am afraid my idea is based on your past comment > If you can't handle a specific I/O request failure gracefully, you should > mark the driver as dead, don't do any more I/Os to the disk or cache > device and return -EIO on all incoming requests. > > Always think that I/O failures can happen because of connection problems, > not data corruption problems - for example, a disk cable can go loose, a > network may lose connectivity, etc. In these cases, it is best to stop > doing any I/O at all and let the user resolve the situation. 1) In failure, mark the driver dead - set `blockup` to 1 in my case - and returning -EIO on all incoming requests. Yes. 2) And wait for the user resolve the situation - returning -EIO until admin turns `blockup` to 0 after checkup in my case - . Yes. Did you mean we should not provide any way to recover the system because admin may not be able to reach the switch? writeboost module autonomously checking the device in problem recovered should be implemented? Retry submitting I/O to the device and find the device is recovered on I/O success is a solution and I have implemented it. I/O retry doesn't destroy any consistency in writeboost; sooner or later it can not be able to accept writes any more because of lack of RAM buffer which can be reused after I/O success to cache device. Akira -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
