On Fri, Jul 06, 2018 at 07:51:30AM -0600, Jens Axboe wrote: > On 7/3/18 11:35 AM, Mikulas Patocka wrote: > > Discarding can take very long time for some device mapper targets, this > > patch makes it possible to kill a process that issues the BLKDISCARD > > ioctl. > > > > Note that some filesystems call blkdev_issue_discard or > > __blkdev_issue_discard directly, they may not be prepared to handle the > > failure, so this patch introduces a flag BLKDEV_DISCARD_KILLABLE that is > > only set when the discard is initiated by an ioctl. > > This might be cleaner as a regular request flag, since killable can > apply to other types of IO as well - like readahead. He doesn't actually make the request killable, but just the synchronous submission loop. I have an actually killable version of blk_execute_rq on m short-term todo plate as we want it for the nvme passthrough ioctls. That might be useful for discard as well.
