On 8/31/2023 7:39 PM, Vincent Fu wrote: > I think the metadata size check is too strict. Commands where the metadata size > is too small should result in errors but when the metadata size is larger than > needed they should still go through. Indeed. I will fold that change in the next version. > In any case, I tested this patch on a QEMU NVMe device (which supports PI by > default). > > I formatted the device with a 512+16 lbaf with a separate buffer for metadata: > > nvme format /dev/ng0n1 -m 0 -i 1 -p 0 --lbaf 2 --force > > Using the latest fio I wrote some data to it: > > ./fio --name=difdix --ioengine=io_uring_cmd --cmd_type=nvme \ > --filename=/dev/ng0n1 --rw=write --bs=512 --md_per_io_size=16 --pi_act=1 \ > --pi_chk=APPTAG --apptag=0x8888 --apptag_mask=0xFFFF --number_ios=128 > > Then I wrote a small program to read 4096 bytes from the device with only a > 16-byte (instead of 64-byte) metadata buffer. Without this patch the kernel > crashes. With the patch the read fails with an error message in the kernel log. > > Tested-by: Vincent Fu <vincent.fu@xxxxxxxxxxx> Thanks.
