On 19/07/07, M. Warner Losh <[EMAIL PROTECTED]> wrote: [...]
The best one can do without massive buffer cache work is what firewire does: it has one attachment to handle all umass devices. When the device goes away, it pauses all operations to that device. If the device comes back, it resumes the I/O . If the device never comes back, then the I/O never finishes.
Is this safe? I don't know where locking occurs in this case, but if locking occurs on a very low level it's potentially dangerous. If a device is removed (either on purpose or by accident) the kernel can't determine the state of the filesystem anymore. So the user could plug the device into another machine, start some write operations on the device, and put it back into the FreeBSD machine. This wouldn't know anything about the changes done, and just flush its buffer, probably using blocks that have been filled previously. It's a pity that FreeBSD can't handle these situations. Since no one here on this list has enough money to get development on the road, maybe we could try collecting money? Everyone interested in seeing this issue fixed offers the amount of money he/she likes to spend... I guess for a "Summer of Code" project this issue would be to big to fix, wouldn't it? _______________________________________________ [email protected] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "[EMAIL PROTECTED]"
