On 4/15/25 9:54 PM, Ming Lei wrote: > Hello, > > Patch 1st ~ 7th simplifies & improves IO canceling when ublk server daemon > is exiting by taking two stage canceling: > > - canceling active uring_cmd from its cancel function > > - move inflight requests aborting into ublk char device release handler > > With this way, implementation is simplified a lot, meantime ub->mutex is > not required before queue becomes quiesced, so forward progress is > guaranteed. > > This approach & main implementation is from Uday's patch of > "improve detection and handling of ublk server exit". > > The last patch is selftest code for showing the improvement ublk server > exit, 30sec timeout is avoided, which depends on patchset of > "[PATCH V2 00/13] selftests: ublk: test cleanup & add more tests"[1]. > > [1] https://lore.kernel.org/linux-block/20250412023035.2649275-1-ming.lei@xxxxxxxxxx/T/#medbf7024e57beaf1c53e4cef6e076421463839d0 > > Pass both ublk kernel selftests and ublksrv 'make test T=generic'. Looks good to me - what are we targeting with this patchset? I think an argument could be made for 6.15, curious what you're thinking? -- Jens Axboe
