linux-kernel - Re: [RFC][PATCH] fix for async scsi scan sysfs problem (resend)

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20070421135956.GA22873@korben.rdu.redhat.com>
Date:	Sat, 21 Apr 2007 09:59:56 -0400
From:	Josef Bacik <jwhiter@...hat.com>
To:	Andrew Morton <akpm@...ux-foundation.org>
Cc:	Josef Bacik <jwhiter@...hat.com>,
	James Bottomley <James.Bottomley@...elEye.com>,
	linux-scsi@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [RFC][PATCH] fix for async scsi scan sysfs problem (resend)

On Sat, Apr 21, 2007 at 12:23:45AM -0700, Andrew Morton wrote:
> On Thu, 19 Apr 2007 11:06:56 -0400 Josef Bacik <jwhiter@...hat.com> wrote:
> 
> > On Thu, Apr 19, 2007 at 10:02:36AM -0400, James Bottomley wrote:
> > > On Thu, 2007-04-19 at 09:25 -0400, Josef Bacik wrote:
> > > > Looking through everything I came to the conclusion that we don't really need
> > > > the scsi_sysfs_add_devices in scsi_finish_async_scan, which gets run everytime
> > > > we do a do_scan_async.  In doing the scanning, if we come upon anything we will
> > > > already be registering the device with sysfs so the scsi_sysfs_add_devices step
> > > > is kind of useless.
> > > 
> > > Unfortunately, it isn't.  The registration step while scanning is at the
> > > end of scsi_add_lun():
> > > 
> > > 
> > > 	if (!async && scsi_sysfs_add_sdev(sdev) != 0)
> > > 		return SCSI_SCAN_NO_RESPONSE;
> > > 
> > > 	return SCSI_SCAN_LUN_PRESENT;
> > > 
> > > The !async should mean that the addition *only* occurs for the non async
> > > scan case ... if you remove the post async scan add, we'll lose devices.
> > > 
> > > >   I tested this and it worked fine on my UP box (where the
> > > > problem was happening) and my SMP box (where the problem wasn't happening).  Now
> > > > I'm not entirely sure if this is correct, but I'm attaching the patch that I
> > > > used to fix it for me, please point out if I've done something wrong or if there
> > > > is a different way this needs to be fixed.  Thank you,
> > > 
> > > Could you add some debugging first to see if we're actually adding the
> > > device twice (and also, if we are, what the value of the async is).
> > >
> > 
> > Sorry I should have put that in the original post, I added debugging to
> > kobject_add to check to see if we were adding something twice, thats how I
> > figured out who was doing it
> > 
> > kobject rport-3:0-0: registering. parent: host3, set: devices
> > kobject rport-3:0-0: registering. parent: fc_remote_ports, set: class_obj
> > kobject target3:0:0: registering. parent: rport-3:0-0, set: devices
> > kobject rport-3:0-1: registering. parent: host3, set: devices
> > kobject rport-3:0-1: registering. parent: fc_remote_ports, set: class_obj
> > kobject target3:0:0: registering. parent: fc_transport, set: class_obj
> > kobject rport-3:0-2: registering. parent: host3, set: devices
> > kobject rport-3:0-2: registering. parent: fc_remote_ports, set: class_obj
> > kobject rport-3:0-3: registering. parent: host3, set: devices
> > kobject rport-3:0-3: registering. parent: fc_remote_ports, set: class_obj
> > kobject rport-3:0-4: registering. parent: host3, set: devices
> > kobject rport-3:0-4: registering. parent: fc_remote_ports, set: class_obj
> > kobject rport-3:0-5: registering. parent: host3, set: devices
> > kobject rport-3:0-5: registering. parent: fc_remote_ports, set: class_obj
> > kobject rport-3:0-6: registering. parent: host3, set: devices
> > kobject rport-3:0-6: registering. parent: fc_remote_ports, set: class_obj
> > kobject rport-3:0-7: registering. parent: host3, set: devices
> > kobject rport-3:0-7: registering. parent: fc_remote_ports, set: class_obj
> > >> kobject 3:0:0:0: registering. parent: target3:0:0, set: devices
> > kobject 3:0:0:0: registering. parent: scsi_device, set: class_obj
> > scsi 3:0:0:0: Direct-Access     IBM 1742-900         0520 PQ: 0 ANSI: 3
> > >> kobject 3:0:0:0: registering. parent: target3:0:0, set: devices
> > kobject_add failed for 3:0:0:0 with -EEXIST, don't try to register things with 
> > the same name in the same directory.
> > 
> > Async in the first case is set and in the second case it isn't set.  Thank you,
> > 
> 
> So.... do we now know what is causing this failure?

Yes, the qla init stuff kicks off the async scanning, and then continues to
initialize, and registers itself with the scsi fc stuff, which makes a workqueue
to register the fc port.  The problem with this is the async scanning and the fc
port registration stuff runs back to back on my UP computer, so the scsi async
stuff does the sysfs registration and then right after that the scsi fc stuff
goes through and does its own scsi scan without async set, which does the sysfs
registration as well, and then kobject complains about adding the same thing
twice.  Taking out the async check in scsi_add_lun would probably work, but this
really isn't my area of expertise and I have a feeling thats kind of defeating
the purpose of the async scsi scanning.

Josef
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/