misc/144695: race condition in mounting a root-fs on an external usb-disk

[Garrett Cooper]

The following reply was made to PR kern/144695; it has been noted by GNATS.

From: Garrett Cooper <yanefbsd at gmail.com>
To: Torsten <tkalix at bredex.de>
Cc: freebsd-gnats-submit at freebsd.org
Subject: Re: misc/144695: race condition in mounting a root-fs on an external 
	usb-disk
Date: Fri, 12 Mar 2010 14:18:44 -0800

 On Fri, Mar 12, 2010 at 6:27 AM, Torsten <tkalix at bredex.de> wrote:
 >
 >>Number: =A0 =A0 =A0 =A0 144695
 >>Category: =A0 =A0 =A0 misc
 >>Synopsis: =A0 =A0 =A0 race condition in mounting a root-fs on an external=
  usb-disk
 >>Confidential: =A0 no
 >>Severity: =A0 =A0 =A0 non-critical
 >>Priority: =A0 =A0 =A0 low
 >>Responsible: =A0 =A0freebsd-bugs
 >>State: =A0 =A0 =A0 =A0 =A0open
 >>Quarter:
 >>Keywords:
 >>Date-Required:
 >>Class: =A0 =A0 =A0 =A0 =A0sw-bug
 >>Submitter-Id: =A0 current-users
 >>Arrival-Date: =A0 Fri Mar 12 14:30:03 UTC 2010
 >>Closed-Date:
 >>Last-Modified:
 >>Originator: =A0 =A0 Torsten
 >>Release: =A0 =A0 =A0 =A0freebsd 8.0 release
 >>Organization:
 > n/a
 >>Environment:
 >>Description:
 > Hello!
 >
 > I have installed FreeBSD 8 on my external usb-harddisk without any proble=
 ms. Just I cannot boot into it, but I always receive a mountroot-prompt.
 > Reading the news / archives I found a problem in the mountroot taking pla=
 ce too early -- before all of the usb-probe and device-setup is fully done.
 > There was (?) no really solution available, so I have made a small change=
  that caused everything to work -- but one may see this more a work-around =
 than rather a real fix.
 > The main idea is to put in an un-conditional wait at the beginning of vfs=
 _mountroot. And, of course make this configurable. The acutal change is jus=
 t a few lines that should not harm anything else.
 > For my system I have started with the default of 5000ms, but I needed to =
 go to 15000ms to mount my root-fs safely.
 > Please find attached the source-diff.
 >>How-To-Repeat:
 >
 >>Fix:
 > FreeBSD unixrest 8.0-RELEASE FreeBSD 8.0-RELEASE #0: Fri Mar 12 14:02:27 =
 CET 2010 =A0 =A0 root at unixrest:/usr/src/sys/i386/compile/BXGENERIC =A0i386
 >
 >
 > Patch attached with submission follows:
 >
 > --- vfs_mount.c.orig =A0 =A02010-03-12 13:02:33.000000000 +0100
 > +++ vfs_mount.c 2010-03-12 13:56:04.000000000 +0100
 > @@ -1648,6 +1648,12 @@
 >
 > =A0 =A0 =A0 =A0options =3D NULL;
 >
 > + =A0 =A0 =A0 =A0static int mountroot_wait_delay =3D 5000;
 > + =A0 =A0 =A0 =A0TUNABLE_INT_FETCH("hw.mountroot_wait_delay", &mountroot_=
 wait_delay);
 > + =A0 =A0 =A0 int mysec =3D mountroot_wait_delay / hz;
 > + =A0 =A0 =A0 printf("Waiting %d seconds for devices to settle.\n", mysec=
 );
 > + =A0 =A0 =A0 =A0pause("Waiting for devices to settle", mountroot_wait_de=
 lay);
 > +
 > =A0 =A0 =A0 =A0root_mount_prepare();
 >
 > =A0 =A0 =A0 =A0mount_zone =3D uma_zcreate("Mountpoints", sizeof(struct mo=
 unt),
 > @@ -2492,3 +2498,4 @@
 > =A0 =A0 =A0 =A0error =3D kernel_mount(ma, flags);
 > =A0 =A0 =A0 =A0return (error);
 > =A0}
 > +
 
 1. This was misfiled (should be under kern).
 2. This is a well-known issue (devices in general not being available
 by the time mount of mountroot is attempted) that isn't trivial to
 resolve.
 3. mav@ has a more scalable solution offered here that's been
 committed into CURRENT in the past couple months:
 http://www.mail-archive.com/freebsd-stable@freebsd.org/msg107696.html
 . Look for kern.cam.boot_delay.
 
 HTH,
 -Garrett