Versions:

kernel-4.18.0-58.el8.x86_64
qemu-kvm-2.12.0-57.module+el8+2683+02b3b955




1. Configure the multipath device on host 

# multipath -ll
mpatha (36001405c6eefcbb279645ef9d02f3f91) dm-3 LIO-ORG,stor0
size=40G features='0' hwhandler='1 alua' wp=rw
|-+- policy='service-time 0' prio=50 status=active
| `- 15:0:0:0 sdb 8:16 active ready running
`-+- policy='service-time 0' prio=50 status=enabled
  `- 16:0:0:0 sdc 8:32 active ready running

2. passthrough it withount "werror=stop,rerror=stop"

    -device virtio-scsi-pci,id=scsi1,bus=pci.0 \
    -drive file=/dev/sdb,format=raw,if=none,id=drive-scsi0-0-0-0,cache=none,aio=threads \
    -device scsi-block,bus=scsi1.0,drive=drive-scsi0-0-0-0,id=scsi0-0-0-0 \

3. create partition on passthrough disk


after step 3, qemu core dump



if add "werror=stop,rerror=stop" to cmd lines, qemu does not core dump but failed to create partition. It will prompt "the requested resource is in use" when create partition.


This issue block bug [1], so set status to high.

[1] Bug 1633179 - Hit io-error when run validate SCSI-3 Persistent Reservation test on win2012 guest

Rich Jones reported it as a regression from 40dce4ee61[1] in Bug #1650975, but then Xueqiang Wei reproduced it with QEMU-2.12 in RHEL8. We probably have different issues here.

Xueqiang Wei: why do you think your problem (comment #1) is the same as the one reported by Rich?

The commit in question is not in RHEL-8, it's in QEMU-3.1 (RHEL-8 virt fast train).

1:
commit 40dce4ee61c68395f6d463fae792f61b7c003bce
Author: Paolo Bonzini <pbonzini>
Date:   Sat Oct 13 11:52:34 2018 +0200

    scsi-disk: fix rerror/werror=ignore
    
    rerror=ignore was returning true from scsi_handle_rw_error but the callers were not
    calling scsi_req_complete when rerror=ignore returns true (this is the correct thing
    to do when true is returned after executing a passthrough command).  Fix this by
    calling it in scsi_handle_rw_error.
    
    Signed-off-by: Paolo Bonzini <pbonzini>

(In reply to Ademar Reis from comment #3)
> Rich Jones reported it as a regression from 40dce4ee61[1] in Bug #1650975,
> but then Xueqiang Wei reproduced it with QEMU-2.12 in RHEL8. We probably
> have different issues here.
> 
> Xueqiang Wei: why do you think your problem (comment #1) is the same as the
> one reported by Rich?
> 
> The commit in question is not in RHEL-8, it's in QEMU-3.1 (RHEL-8 virt fast
> train).
> 
> 1:
> commit 40dce4ee61c68395f6d463fae792f61b7c003bce
> Author: Paolo Bonzini <pbonzini>
> Date:   Sat Oct 13 11:52:34 2018 +0200
> 
>     scsi-disk: fix rerror/werror=ignore
>     
>     rerror=ignore was returning true from scsi_handle_rw_error but the
> callers were not
>     calling scsi_req_complete when rerror=ignore returns true (this is the
> correct thing
>     to do when true is returned after executing a passthrough command).  Fix
> this by
>     calling it in scsi_handle_rw_error.
>     
>     Signed-off-by: Paolo Bonzini <pbonzini>


The output and core dump file log are same.

QEMU 2.12.0 monitor - type 'help' for more information
(qemu) qemu-kvm: hw/scsi/scsi-bus.c:1374: scsi_req_complete: Assertion `req->status == -1' failed.


(gdb) bt
#0  0x00007f3e4bcf193f in raise () at /lib64/libc.so.6
#1  0x00007f3e4bcdbc95 in abort () at /lib64/libc.so.6
#2  0x00007f3e4bcdbb69 in _nl_load_domain.cold.0 () at /lib64/libc.so.6
#3  0x00007f3e4bce9df6 in .annobin_assert.c_end () at /lib64/libc.so.6
#4  0x000055a83bea7040 in scsi_req_complete
    (req=<optimized out>, status=<optimized out>) at hw/scsi/scsi-bus.c:1374
#5  0x000055a83bea138a in scsi_handle_rw_error
    (acct_failed=false, error=0, r=0x55a83e852880) at hw/scsi/scsi-disk.c:471
#6  0x000055a83bea138a in scsi_disk_req_check_error
    (r=0x55a83e852880, ret=<optimized out>, acct_failed=<optimized out>)
    at hw/scsi/scsi-disk.c:196
#7  0x000055a83bea2225 in scsi_dma_complete_noio
    (r=0x55a83e852880, ret=<optimized out>) at hw/scsi/scsi-disk.c:269
#8  0x000055a83bea2303 in scsi_dma_complete (opaque=0x55a83e852880, ret=0)
    at hw/scsi/scsi-disk.c:300
#9  0x000055a83bdea4d2 in dma_complete (ret=0, dbs=0x7f3e38091ee0)
    at dma-helpers.c:116
#10 0x000055a83bdea4d2 in dma_blk_cb (opaque=0x7f3e38091ee0, ret=0)
    at dma-helpers.c:138
#11 0x000055a83bf69a9e in blk_aio_complete (acb=0x7f3e30000f20)
    at block/block-backend.c:1346
#12 0x000055a83c0080f3 in coroutine_trampoline
    (i0=<optimized out>, i1=<optimized out>) at util/coroutine-ucontext.c:116
--Type <RET> for more, q to quit, c to continue without paging--
#13 0x00007f3e4bd07600 in __start_context () at /lib64/libc.so.6
#14 0x00007f3e22ffbc10 in  ()
#15 0x0000000000000000 in  ()

There is no fix yet.  We can delay it.

The bug has been fixed for a long time in RHEL AV, so moving the bug for simplicity to RHEL qemu-kvm.

It's fixed by upstream commit 1c7f618f689b0b5b6bbed23a7c159e7dad7b996f.

Reassigning (with ack from Paolo).

Fixed in current release, right?