| Message ID | 20221019081825.20794-1-yangxingui@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | scsi: sd: Update dix config everytime sd_revalidate_disk is called | expand |
Xingui, > When the host protection capabilities are 0x77 and a DIF disk is > connected, the DIX and DIF of the disk are default enabled. Then if > that DIF disk is reformatted as a non-DIF format, per the currently > flow, the DIX is kept enabled which is not correct DIX is orthogonal to DIF and a host property. DIX should not be disabled just because the disk is no longer formatted with protection information. That said, your patch is probably fine. I'll take a closer look tomorrow.
On 2022/10/20 8:18, Martin K. Petersen wrote: > Xingui, > >> When the host protection capabilities are 0x77 and a DIF disk is >> connected, the DIX and DIF of the disk are default enabled. Then if >> that DIF disk is reformatted as a non-DIF format, per the currently >> flow, the DIX is kept enabled which is not correct > DIX is orthogonal to DIF and a host property. DIX should not be disabled > just because the disk is no longer formatted with protection > information. As the host protection capabilities are 0x77 and not support DIX between OS and HBA only. If write command is delivered to a non-DIF disk with DIX format this time, I/Os will be failed. DIX status should be updated when the disk DIF type changes. > > That said, your patch is probably fine. I'll take a closer look > tomorrow. ok, thanks. Xingui >
On 2022/10/20 8:18, Martin K. Petersen wrote: > > Xingui, > >> When the host protection capabilities are 0x77 and a DIF disk is >> connected, the DIX and DIF of the disk are default enabled. Then if >> that DIF disk is reformatted as a non-DIF format, per the currently >> flow, the DIX is kept enabled which is not correct > > DIX is orthogonal to DIF and a host property. DIX should not be disabled > just because the disk is no longer formatted with protection > information. > > That said, your patch is probably fine. I'll take a closer look > tomorrow. > Friendly ping... Do you have other suggestions? Thanks, Xingui
diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c index eb76ba055021..3844f469b6be 100644 --- a/drivers/scsi/sd.c +++ b/drivers/scsi/sd.c @@ -2199,9 +2199,6 @@ static void sd_config_protection(struct scsi_disk *sdkp) { struct scsi_device *sdp = sdkp->device; - if (!sdkp->first_scan) - return; - sd_dif_config_host(sdkp); if (!sdkp->protection_type) diff --git a/drivers/scsi/sd_dif.c b/drivers/scsi/sd_dif.c index 968993ee6d5d..78db8d85f97e 100644 --- a/drivers/scsi/sd_dif.c +++ b/drivers/scsi/sd_dif.c @@ -39,8 +39,10 @@ void sd_dif_config_host(struct scsi_disk *sdkp) dif = 0; dix = 1; } - if (!dix) + if (!dix) { + blk_integrity_unregister(disk); return; + } memset(&bi, 0, sizeof(bi));
When the host protection capabilities are 0x77 and a DIF disk is connected, the DIX and DIF of the disk are default enabled. Then if that DIF disk is reformatted as a non-DIF format, per the currently flow, the DIX is kept enabled which is not correct, which will cause the following errors when accessing the non-DIF disk: [root@localhost ~]# lsscsi -p [7:0:5:0] disk xxx /dev/sdc DIF/Type3 T10-DIF-TYPE3-CRC [root@localhost ~]# sg_format -F -s 512 /dev/sdc [root@localhost ~]# lsscsi -p [7:0:5:0] disk xxx /dev/sdc - T10-DIF-TYPE3-CRC [142829.032340] hisi_sas_v3_hw 0000:b4:04.0: erroneous completion iptt=2375 task=00000000bea0970c dev id=5 direct-attached phy4 addr=51c20dbaf642a000 CQ hdr: 0x1023 0x50947 0x0 0x20000 Error info: 0x0 0x0 0x4 0x0 [142829.073883] sas: Enter sas_scsi_recover_host busy: 1 failed: 1 [142829.079783] sas: sas_scsi_find_task: aborting task 0x00000000bea0970c [142829.102342] sas: Internal abort: task to dev 51c20dbaf642a000 response: 0x0 status 0x5 [142829.110319] sas: sas_eh_handle_sas_errors: task 0x00000000bea0970c is done [142829.117275] sd 7:0:5:0: [sdc] tag#2375 UNKNOWN(0x2003) Result: hostbyte=0x05 driverbyte=DRIVER_OK cmd_age=0s [142829.127171] sd 7:0:5:0: [sdc] tag#2375 CDB: opcode=0x2a 2a 00 00 00 00 00 00 00 08 00 [142829.135059] I/O error, dev sdc, sector 0 op 0x1:(WRITE) flags 0x18800 phys_seg 1 prio class 2 On the contrary, when a non-DIF disk is connected and formatted as a DIF disk, it is found that DIX is not enabled. Operation logs as follows: [root@localhost ~]# lsscsi -p [7:0:2:0] disk xxx /dev/sdc - none [root@localhost ~]# sg_format --format --fmtpinfo=3 --pfu=1 /dev/sdc [root@localhost ~]# lsscsi -p [7:0:2:0] disk xxx /dev/sdc DIF/Type3 none This is because dix config is only updated when the first time a disk is added, in this patch, we fix the issue by with changes: 1. Remove check first_scan when call sd_config_protection(). 2. Unregister block integrity profile after DIX becomes to 0. Signed-off-by: Xingui Yang <yangxingui@huawei.com> --- drivers/scsi/sd.c | 3 --- drivers/scsi/sd_dif.c | 4 +++- 2 files changed, 3 insertions(+), 4 deletions(-)