md disk fault communication code

md disk fault communication code

[Sonu a]

when disk is removed with out mdadm as I see from the stack below the
communication reaching the md driver.

dump_stack+0x49/0x5e
md_error+0x50/0x110 [md_mod]
state_store+0x43/0x300 [md_mod]
rdev_attr_store+0xad/0xd0 [md_mod]
? sysfs_write_file+0x62/0x1c0
sysfs_write_file+0x138/0x1c0
vfs_write+0xc0/0x1e0
SyS_write+0x5a/0xa0
? __audit_syscall_exit+0x246/0x2f0
system_call_fastpath+0x16/0x1b

could someone point me to the code which is monitoring scsi disks
status and thus calling md driver sysfs interface accordingly ?

Thx.

Re: md disk fault communication code

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 1194 bytes --]

On Fri, 18 Apr 2014 13:38:58 +0800 Sonu a <p10sonu@gmail.com> wrote:

> when disk is removed with out mdadm as I see from the stack below the
> communication reaching the md driver.
> 
> dump_stack+0x49/0x5e
> md_error+0x50/0x110 [md_mod]
> state_store+0x43/0x300 [md_mod]
> rdev_attr_store+0xad/0xd0 [md_mod]
> ? sysfs_write_file+0x62/0x1c0
> sysfs_write_file+0x138/0x1c0
> vfs_write+0xc0/0x1e0
> SyS_write+0x5a/0xa0
> ? __audit_syscall_exit+0x246/0x2f0
> system_call_fastpath+0x16/0x1b
> 
> could someone point me to the code which is monitoring scsi disks
> status and thus calling md driver sysfs interface accordingly ?

I think you ask asking how md_error gets called when a SCSI device fails,
having already discovered how it is called when you explicitly write to a
sysfs file.

Nothing monitors the scsi disks.  md only discovers failure if it sends a
request to a disk, and the request signals an error.  If you search for
'bi_end_io', functions assigned to this field are called when a request
finishes.  Those functions might call md_error if the request failed, or they
might schedule some other handling first to try to correct the error.

NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Re: md disk fault communication code

[Sonu a]

Yes it does when there is IO failure But.

But my question was when disk fail silently with out IO as show below.

The md sysfs interface /sys/block/mdY/md/dev-sdX/state is written with
faulty when sd corresponding disk is deleted with..

echo 1 >  /sys/block/sdc/device/delete

kernel: [21853.981735] sd 2:0:0:0: [sdc] Synchronizing SCSI cache
kernel: [21854.049967] md: md0 still in use.
kernel: [21854.051201] md/raid1:md0: Disk failure on sdc, disabling device.
kernel: [21854.051201] md/raid1:md0: Operation continuing on 1 devices.
kernel: [21854.308355] sd 2:0:0:0: [sdc] Stopping disk
kernel: [21854.415122] ata3.00: disabled
kernel: [21854.467540] md: unbind<sdc>
kernel: [21854.467544] md: export_rdev(sdc)

earlier stack dump which shows the sysfs write interface

there has to be code monitoring block disk state, and propagating that
state to the md ?

Thx.

On Fri, Apr 18, 2014 at 2:13 PM, NeilBrown <neilb@suse.de> wrote:
> On Fri, 18 Apr 2014 13:38:58 +0800 Sonu a <p10sonu@gmail.com> wrote:
>
>> when disk is removed with out mdadm as I see from the stack below the
>> communication reaching the md driver.
>>
>> dump_stack+0x49/0x5e
>> md_error+0x50/0x110 [md_mod]
>> state_store+0x43/0x300 [md_mod]
>> rdev_attr_store+0xad/0xd0 [md_mod]
>> ? sysfs_write_file+0x62/0x1c0
>> sysfs_write_file+0x138/0x1c0
>> vfs_write+0xc0/0x1e0
>> SyS_write+0x5a/0xa0
>> ? __audit_syscall_exit+0x246/0x2f0
>> system_call_fastpath+0x16/0x1b
>>
>> could someone point me to the code which is monitoring scsi disks
>> status and thus calling md driver sysfs interface accordingly ?
>
> I think you ask asking how md_error gets called when a SCSI device fails,
> having already discovered how it is called when you explicitly write to a
> sysfs file.
>
> Nothing monitors the scsi disks.  md only discovers failure if it sends a
> request to a disk, and the request signals an error.  If you search for
> 'bi_end_io', functions assigned to this field are called when a request
> finishes.  Those functions might call md_error if the request failed, or they
> might schedule some other handling first to try to correct the error.
>
> NeilBrown

Re: md disk fault communication code

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 2833 bytes --]

On Fri, 18 Apr 2014 14:47:06 +0800 Sonu a <p10sonu@gmail.com> wrote:

> Yes it does when there is IO failure But.
> 
> But my question was when disk fail silently with out IO as show below.
> 
> The md sysfs interface /sys/block/mdY/md/dev-sdX/state is written with
> faulty when sd corresponding disk is deleted with..
> 
> echo 1 >  /sys/block/sdc/device/delete
> 
> kernel: [21853.981735] sd 2:0:0:0: [sdc] Synchronizing SCSI cache
> kernel: [21854.049967] md: md0 still in use.
> kernel: [21854.051201] md/raid1:md0: Disk failure on sdc, disabling device.
> kernel: [21854.051201] md/raid1:md0: Operation continuing on 1 devices.
> kernel: [21854.308355] sd 2:0:0:0: [sdc] Stopping disk
> kernel: [21854.415122] ata3.00: disabled
> kernel: [21854.467540] md: unbind<sdc>
> kernel: [21854.467544] md: export_rdev(sdc)
> 
> earlier stack dump which shows the sysfs write interface
> 
> there has to be code monitoring block disk state, and propagating that
> state to the md ?

I understand your question now.

This is handled by used.  /usr/lib/udev/rules.d/64-md-raid-assembly.rules or
some file name like that contains a line like

ACTION=="remove", ENV{ID_PATH}!="?*", RUN+="/sbin/mdadm -If $name"

so when the device is removed, udev runs "mdadm -If /dev/devicename".
mdadm finds which array this device is in, marks it as faulty via sysfs, and
then removes the device from the array if it can.

NeilBrown


> 
> Thx.
> 
> On Fri, Apr 18, 2014 at 2:13 PM, NeilBrown <neilb@suse.de> wrote:
> > On Fri, 18 Apr 2014 13:38:58 +0800 Sonu a <p10sonu@gmail.com> wrote:
> >
> >> when disk is removed with out mdadm as I see from the stack below the
> >> communication reaching the md driver.
> >>
> >> dump_stack+0x49/0x5e
> >> md_error+0x50/0x110 [md_mod]
> >> state_store+0x43/0x300 [md_mod]
> >> rdev_attr_store+0xad/0xd0 [md_mod]
> >> ? sysfs_write_file+0x62/0x1c0
> >> sysfs_write_file+0x138/0x1c0
> >> vfs_write+0xc0/0x1e0
> >> SyS_write+0x5a/0xa0
> >> ? __audit_syscall_exit+0x246/0x2f0
> >> system_call_fastpath+0x16/0x1b
> >>
> >> could someone point me to the code which is monitoring scsi disks
> >> status and thus calling md driver sysfs interface accordingly ?
> >
> > I think you ask asking how md_error gets called when a SCSI device fails,
> > having already discovered how it is called when you explicitly write to a
> > sysfs file.
> >
> > Nothing monitors the scsi disks.  md only discovers failure if it sends a
> > request to a disk, and the request signals an error.  If you search for
> > 'bi_end_io', functions assigned to this field are called when a request
> > finishes.  Those functions might call md_error if the request failed, or they
> > might schedule some other handling first to try to correct the error.
> >
> > NeilBrown


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]