Hi Ming,

You are right.
I moved the drive to onboard SATA controller and the problem disappeared.

As mpt3sas driver is from ELRepo (RedHat dropped support for this
controller in RHEL 8.x) I don't know if I should continue it here or
contact somebody from ELRepo?

Regards,
Olaf


pon., 21 lut 2022 o 15:45 Ming Lei [off-list ref] napisał(a):

On Mon, Feb 21, 2022 at 02:09:12PM +0100, Lukas Czerner wrote:

quoted

Hi,

the problem is definitelly not in util-linux. In kernel there are checks
in place that would prevent proceeding with out of range BLKDISCARD ioctl,
but that's not what we hit here.

In the logs below you can see that the actual discard request failed,
but it appears to be well within the device range. I don't know what is
going on, maybe someone in the linux-block have a clue (adding to cc).

Meanwhile please let us know what kernel version do you have and provide
a blkparse output of the blkdiscard run. You can do this for example

blktrace -a discard -d /dev/sdb -o - | \
blkparse -o output -f "%D %2c %8s %5T.%9t %6p %2a %3d %10S + %10U (%4e) [%C]\n" -i -

then run the blkdiscard and see the content of output file.

Thanks!
-Lukas



On Mon, Feb 21, 2022 at 01:34:57PM +0100, Olaf Fraczyk wrote:

quoted

Hello,

I had to put the disk in use, and I needed it in MBR format, so I can't
create GPT now.

Anyway, the reported size seems to be OK.

I have created 3rd partition to go till the end of the disk, as below:

Device     Boot      Start        End    Sectors   Size Id Type
/dev/sdb1             2048    4196351    4194304     2G fd Linux raid
autodetect
/dev/sdb2          4196352 1874855935 1870659584   892G fd Linux raid
autodetect
/dev/sdb3       1874855936 1875385007     529072 258.3M 83 Linux

I can fill it to the last sector using dd without problems:

[root@vh3 ~]# dd if=/dev/zero of=/dev/sdb3 bs=1024 count=264536
264536+0 records in
264536+0 records out
270884864 bytes (271 MB, 258 MiB) copied, 4.81622 s, 56.2 MB/s

When I do blkdiscard:

root@vh3 ~]# blkdiscard -l 264536K /dev/sdb3
blkdiscard: /dev/sdb3: BLKDISCARD ioctl failed: Remote I/O error
[root@vh3 ~]# blkdiscard -l 264535K /dev/sdb3
[root@vh3 ~]#

In the /var/log/messages for the failed discard I get:
Feb 21 13:19:52 vh3 kernel: sd 1:0:1:0: [sdb] tag#2227 FAILED Result:
hostbyte=DID_OK driverbyte=DRIVER_SENSE cmd_age=0s
Feb 21 13:19:52 vh3 kernel: sd 1:0:1:0: [sdb] tag#2227 Sense Key : Illegal
Request [current]
Feb 21 13:19:52 vh3 kernel: sd 1:0:1:0: [sdb] tag#2227 Add. Sense: Logical
block address out of range
Feb 21 13:19:52 vh3 kernel: sd 1:0:1:0: [sdb] tag#2227 CDB: Unmap/Read
sub-channel 42 00 00 00 00 00 00 00 18 00
Feb 21 13:19:52 vh3 kernel: blk_update_request: critical target error, dev
sdb, sector 1874855936 op 0x3:(DISCARD) flags 0x800 phys_seg 1 prio class 0

I have the drive on a SAS controller - mpt3sas driver, LSI SAS2008

Looks one target issue, CC linux-scsi and mpt3sas guys.

quoted

quoted

quoted

quoted

I tried to trim entire drive but I get the following error:
[root@vh3 util-linux-2.38-rc1]# ./blkdiscard /dev/sdb
lt-blkdiscard: /dev/sdb: BLKDISCARD ioctl failed: Remote I/O error

I have done strace and I see:
ioctl(3, BLKGETSIZE64, [960197124096])  = 0
ioctl(3, BLKSSZGET, [512])              = 0
ioctl(3, BLKDISCARD, [0, 960197124096]) = -1 EREMOTEIO (Remote I/O error)

When I do the same giving length explicitly I get the same error.

However when I specify the length 512 bytes smaller, it works without a
problem:

ioctl(3, BLKGETSIZE64, [960197124096])  = 0
ioctl(3, BLKSSZGET, [512])              = 0
ioctl(3, BLKDISCARD, [0, 960197123584]) = 0

Thanks,
Ming