Hacker News new | ask | show | jobs
by londons_explore 1550 days ago
You ought to be able to flash the firmware on a mounted drive.

Anyone sane implementing the firmware flashing algorithm will either:

1. Just reject all IO requests from the point of the firmware update onwards. The system will see the drive disappear, but since you are using a journaling filesystem, it will be like you just pulled the power, and no data corruption will happen.

2. Complete all IO requests before the update, then update, then continue requests as before.

3. Save the firmware update, but only apply it at next powerup of the drive.
2 comments
chousuke 1549 days ago
#3 is the correct option, I think, except you update the firmware but don't start using it. The OS should handle powercycling the drive at a convenient moment, at which point it'll load the new firmware.

Netapp filers do this; they update disk firmware online by installing it and then just rebooting the drive. The RAID controller/software handles the momentary interruption with zero issues, but even without redundancy an OS can handle a drive disappearing for a few moments.

link
londons_explore 1549 days ago
I don't think Linux can do this... When hardware is powercycled, that would be a disconnect event followed by a connect event. There is no way to be sure it's the same drive connected, or that it will connect again at all. And on the disconnect event, all the datastructures associated with the device get discarded, so all file handles are invalid and nothings gonna work again...
link
chousuke 1549 days ago
Linux has support for multipath devices, so disks disappearing and re-appearing at runtime definitely isn't a new concept. Making it work with a single disk might take some work though.

If the power cycle is controlled, then the kernel can be reasonably sure that the re-appearing device is the same one. File handles wouldn't really be affected as long as the mounted filesystem can resume IO after the device reappears.

Most likely you'd still want to use a RAID scheme and/or data checksumming as an extra layer to ensure data integrity though.

link
yjftsjthsd-h 1550 days ago
You're basically right, but hardware manufacturers are not known for writing sane firmware and a lot of people have been burned enough times to absolutely never ever trust that reasoning.
link