2.7 (future kernel) wish

2.7 (future kernel) wish

[Jos Hulzink]

Hi,

First of all... Compliments about 2.6.0. It is a superb kernel, with very few 
serious bugs, and for me it runs stable like a rock from the very first 
moment.

As an end user, Linux doesn't give me a good feeling on one particular item 
yet: Error handling. 

What do I mean ? Well... for example: Pull out your USB stick with a mounted 
fs on it. Linux doesn't really seem to like it, got weird problems etc. It 
will survive, sure, but the user got no clue and data are lost for sure. Bad 
sectors on a disk... Linux will pass, but even 2.6.0 went very slow, 
unresponsive when a floppy with bad sectors went into the drive. Many other 
non-critical or solvable problems that are dealt with in a way that makes 
linux survive (most of the times), but not in a way that is neat from the 
user point of view.

It all just doesn't feel like Linux is doing the best it can to "rescue the 
user" when something is going wrong. Technically speaking, it's not only the 
task of the kernel to do so, but for an end user it makes the difference 
between an OS that does its job, and an OS that does its job nicely.

I think it's hard to describe what I mean exactly, but I hope you get the 
feeling. I too know that some of this is not within scope of the kernel (it's 
not the kernels task to tell the user "put back the USB drive or data is 
lost"), but after dealing with broken floppies again, I thought it was time 
to write my feelings to the list.

Best regards, and thanks for the wonderful world of Linux,

Jos

Re: 2.7 (future kernel) wish

[Helge Hafting]

On Tue, Dec 23, 2003 at 11:42:17PM +0100, Jos Hulzink wrote:
> Hi,
> 
> First of all... Compliments about 2.6.0. It is a superb kernel, with very few 
> serious bugs, and for me it runs stable like a rock from the very first 
> moment.
> 
> As an end user, Linux doesn't give me a good feeling on one particular item 
> yet: Error handling. 
> 
> What do I mean ? Well... for example: Pull out your USB stick with a mounted 
> fs on it. 

You aren't supposed to do that.  If you want to pull devices like that,
with no warning, access them in other ways than mounting.  
mtools are nice when you don't want to mount/umount floppies - a
similiar approach should work for usb sticks too.



Helge Hafting

Re: 2.7 (future kernel) wish

[David B. Stevens]

While I agree that the kernel should provide decent error handling and 
reporting I still have to ask questions about what is reasonable.

What does that other OS do when you pull a USB stick out?  What do you 
think the kernel should do?  Why don't the applications operating on the 
data take better care of handling error conditions?

I don't have one here to try, but at some point the (ab)user needs to 
take a bit of the heat for his or her action(s) or lack thereof.

After all you could just reach in your case and rip out the IDE or SCSI 
cables.  Bet that leads to all kinds of stuff (tm).

Cheers,
  Dave


Helge Hafting wrote:

>On Tue, Dec 23, 2003 at 11:42:17PM +0100, Jos Hulzink wrote:
>  
>
>>Hi,
>>
>>First of all... Compliments about 2.6.0. It is a superb kernel, with very few 
>>serious bugs, and for me it runs stable like a rock from the very first 
>>moment.
>>
>>As an end user, Linux doesn't give me a good feeling on one particular item 
>>yet: Error handling. 
>>
>>What do I mean ? Well... for example: Pull out your USB stick with a mounted 
>>fs on it. 
>>    
>>
>
>You aren't supposed to do that.  If you want to pull devices like that,
>with no warning, access them in other ways than mounting.  
>mtools are nice when you don't want to mount/umount floppies - a
>similiar approach should work for usb sticks too.
>
>
>
>Helge Hafting
>-
>To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>Please read the FAQ at  http://www.tux.org/lkml/
>
>  
>

Re: 2.7 (future kernel) wish

[Joshua Schmidlkofer]

On Fri, 2003-12-26 at 15:57, David B. Stevens wrote:
> While I agree that the kernel should provide decent error handling and 
> reporting I still have to ask questions about what is reasonable.
> 
> What does that other OS do when you pull a USB stick out?  What do you 
> think the kernel should do?  Why don't the applications operating on the 
> data take better care of handling error conditions?
> 
> I don't have one here to try, but at some point the (ab)user needs to 
> take a bit of the heat for his or her action(s) or lack thereof.
> 
> After all you could just reach in your case and rip out the IDE or SCSI 
> cables.  Bet that leads to all kinds of stuff (tm).
> 
> Cheers,
>   Dave
> 
> 
> Helge Hafting wrote:
> 
> >On Tue, Dec 23, 2003 at 11:42:17PM +0100, Jos Hulzink wrote:
> >  
> >
> >>Hi,
> >>
> >>First of all... Compliments about 2.6.0. It is a superb kernel, with very few 
> >>serious bugs, and for me it runs stable like a rock from the very first 
> >>moment.
> >>
> >>As an end user, Linux doesn't give me a good feeling on one particular item 
> >>yet: Error handling. 
> >>
> >>What do I mean ? Well... for example: Pull out your USB stick with a mounted 
> >>fs on it. 
> >>    
> >>
> >
> >You aren't supposed to do that.  If you want to pull devices like that,
> >with no warning, access them in other ways than mounting.  
> >mtools are nice when you don't want to mount/umount floppies - a
> >similiar approach should work for usb sticks too.
> >
> >
> >
> >Helge Hafting
> >-
> >To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> >the body of a message to majordomo@vger.kernel.org
> >More majordomo info at  http://vger.kernel.org/majordomo-info.html
> >Please read the FAQ at  http://www.tux.org/lkml/
> >

Sometimes Windows 2k or XP dump (BSOD), or maybe you just get an error. 

-- 
VB programmers ask why no one takes them seriously, 
it's somewhat akin to a McDonalds manager asking employees 
why they don't take their 'career' seriously.

Re: 2.7 (future kernel) wish

[Jim Crilly]

Joshua Schmidlkofer wrote:

>On Fri, 2003-12-26 at 15:57, David B. Stevens wrote:
>  
>
>>While I agree that the kernel should provide decent error handling and 
>>reporting I still have to ask questions about what is reasonable.
>>
>>What does that other OS do when you pull a USB stick out?  What do you 
>>think the kernel should do?  Why don't the applications operating on the 
>>data take better care of handling error conditions?
>>
>>I don't have one here to try, but at some point the (ab)user needs to 
>>take a bit of the heat for his or her action(s) or lack thereof.
>>
>>After all you could just reach in your case and rip out the IDE or SCSI 
>>cables.  Bet that leads to all kinds of stuff (tm).
>>
>>Cheers,
>>  Dave
>>
>>
>>Helge Hafting wrote:
>>
>>    
>>
>>>On Tue, Dec 23, 2003 at 11:42:17PM +0100, Jos Hulzink wrote:
>>> 
>>>
>>>      
>>>
>>>>Hi,
>>>>
>>>>First of all... Compliments about 2.6.0. It is a superb kernel, with very few 
>>>>serious bugs, and for me it runs stable like a rock from the very first 
>>>>moment.
>>>>
>>>>As an end user, Linux doesn't give me a good feeling on one particular item 
>>>>yet: Error handling. 
>>>>
>>>>What do I mean ? Well... for example: Pull out your USB stick with a mounted 
>>>>fs on it. 
>>>>   
>>>>
>>>>        
>>>>
>>>You aren't supposed to do that.  If you want to pull devices like that,
>>>with no warning, access them in other ways than mounting.  
>>>mtools are nice when you don't want to mount/umount floppies - a
>>>similiar approach should work for usb sticks too.
>>>
>>>
>>>
>>>Helge Hafting
>>>-
>>>To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>>>the body of a message to majordomo@vger.kernel.org
>>>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>Please read the FAQ at  http://www.tux.org/lkml/
>>>
>>>      
>>>
>
>Sometimes Windows 2k or XP dump (BSOD), or maybe you just get an error. 
>
>  
>
Generally it just complains that you pulled out the device prematurely, 
I've never seen one give a STOP error from that but I guess a bad driver 
or USB controller could cause anything.

When you insert a device like a USB stick Windows puts a little icon 
next to the clock in the system tray that you're supposed to use to stop 
the device before pulling it, effectively it unmounts and stops (or 
atleast releases the device from) the driver so the device can be 
'safely' removed. I also believe Windows mounts any removable device 
synchronously so that if you do pull it out prematurely the damage done 
is limited.

Jim.

Re: 2.7 (future kernel) wish

[Kevin P. Fleming]

Jim Crilly wrote:

> 'safely' removed. I also believe Windows mounts any removable device 
> synchronously so that if you do pull it out prematurely the damage done 
> is limited.

Nope, when I put stuff onto a CF card via CF-to-USB adapter Windows 
still buffers writes to the media while the user interface goes on about 
its business. The only media that I've ever seen Windows use 
synchronously is old-style floppy disks.

Re: 2.7 (future kernel) wish

[Rob Love]

On Sat, 2003-12-27 at 22:08, Kevin P. Fleming wrote:

> Nope, when I put stuff onto a CF card via CF-to-USB adapter Windows 
> still buffers writes to the media while the user interface goes on about 
> its business. The only media that I've ever seen Windows use 
> synchronously is old-style floppy disks.

Yah, it really would not be smart to do all I/O synchronously.

It might be a neat idea to do the buffer writeback much sooner, but that
would require per-device writeback settings, which is ugh.

	Rob Love

RE: 2.7 (future kernel) wish

[Kevin Krieser]

XP came with an option that I've seen on USB hard drives where it won't
cache writes by default.  It is in Device Manager, and called Optimize
for quick removal.  You can also enable write caching, which requires
the use of the Safe Removal icon.

I don't recall 2000 having this option.

With Linux, I'm just ingrained to umount first.

-----Original Message-----
From: linux-kernel-owner@vger.kernel.org
[mailto:linux-kernel-owner@vger.kernel.org] On Behalf Of Kevin P.
Fleming
Sent: Saturday, December 27, 2003 9:08 PM
To: Jim Crilly
Cc: Joshua Schmidlkofer; David B. Stevens; Helge Hafting; Jos Hulzink;
Linux Kernel Mailing List
Subject: Re: 2.7 (future kernel) wish


Jim Crilly wrote:

> 'safely' removed. I also believe Windows mounts any removable device
> synchronously so that if you do pull it out prematurely the damage
done 
> is limited.

Nope, when I put stuff onto a CF card via CF-to-USB adapter Windows 
still buffers writes to the media while the user interface goes on about

its business. The only media that I've ever seen Windows use 
synchronously is old-style floppy disks.

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

Re: 2.7 (future kernel) wish

[Gaël Le Mignot]

Hello Kevin!

Sun, 28 Dec 2003 05:17:11 -0600, you wrote: 

 > XP came with an option that I've seen on USB hard drives where it won't
 > cache writes by default.  It is in Device Manager, and called Optimize
 > for quick removal.  You can also enable write caching, which requires
 > the use of the Safe Removal icon.

 > I don't recall 2000 having this option.

 > With Linux, I'm just ingrained to umount first.

Or with  Linux, you can  use the "sync"  mount option (or  dirsync) to
force direct write of data.

-- 
Gael Le Mignot "Kilobug" - kilobug@nerim.net - http://kilobug.free.fr
GSM         : 06.71.47.18.22 (in France)   ICQ UIN   : 7299959
Fingerprint : 1F2C 9804 7505 79DF 95E6 7323 B66B F67B 7103 C5DA

Member of HurdFr: http://hurdfr.org - The GNU Hurd: http://hurd.gnu.org

Re: 2.7 (future kernel) wish

[Rob Love]

On Sat, 2003-12-27 at 22:03, Jim Crilly wrote:

> Generally it just complains that you pulled out the device prematurely, 
> I've never seen one give a STOP error from that but I guess a bad driver 
> or USB controller could cause anything.

It would be pretty easy to screw things up if you pull out a device in
the middle of use.

> When you insert a device like a USB stick Windows puts a little icon 
> next to the clock in the system tray that you're supposed to use to stop 
> the device before pulling it, effectively it unmounts and stops (or 
> atleast releases the device from) the driver so the device can be 
> 'safely' removed.

This is useful, and something I think we need on the Linux desktop (stay
tuned).

> I also believe Windows mounts any removable device 
> synchronously so that if you do pull it out prematurely the damage done 
> is limited.

Eww, I hope not, that would be excruciatingly slow.  It might adjust the
buffer writeback to be really short (even nearly immediate) but
synchronous I/O is a different story, and much slower.

	Rob Love

Re: 2.7 (future kernel) wish

[Jim Crilly]

Rob Love wrote:
> On Sat, 2003-12-27 at 22:03, Jim Crilly wrote:
> 
> 
>>Generally it just complains that you pulled out the device prematurely, 
>>I've never seen one give a STOP error from that but I guess a bad driver 
>>or USB controller could cause anything.
> 
> 
> It would be pretty easy to screw things up if you pull out a device in
> the middle of use.
> 
> 
>>When you insert a device like a USB stick Windows puts a little icon 
>>next to the clock in the system tray that you're supposed to use to stop 
>>the device before pulling it, effectively it unmounts and stops (or 
>>atleast releases the device from) the driver so the device can be 
>>'safely' removed.
> 
> 
> This is useful, and something I think we need on the Linux desktop (stay
> tuned).
> 

I agree, that's one of the reasons I posted at all. Little things like 
this can make a big difference, even though I've seen a lot of users not 
notice the little icon and have to be told about it.

Maybe when the icon appears have a tool-tip that pops up and says 
something like "your USB device is ready for user at /mnt/usb, click 
here when you're done" or something like that to make it more noticable 
that they shouldn't just yank it.

But I seem to be getting OT for this list...

> 
>>I also believe Windows mounts any removable device 
>>synchronously so that if you do pull it out prematurely the damage done 
>>is limited.
> 
> 
> Eww, I hope not, that would be excruciatingly slow.  It might adjust the
> buffer writeback to be really short (even nearly immediate) but
> synchronous I/O is a different story, and much slower.
> 
> 	Rob Love
> 
> 

Perhaps synchronous was the wrong term =) But it does atleast seem to do 
less buffering for removable devices or I could just be fooled by 
something else slowing it down.

Re: 2.7 (future kernel) wish

[Pat Erley]

On Sat, 27 Dec 2003 22:19:53 -0500
Jim Crilly <jim@why.dont.jablowme.net> wrote:

> >>I also believe Windows mounts any removable device 
> >>synchronously so that if you do pull it out prematurely the damage done 
> >>is limited.
> > 
> > 
> > Eww, I hope not, that would be excruciatingly slow.  It might adjust the
> > buffer writeback to be really short (even nearly immediate) but
> > synchronous I/O is a different story, and much slower.
> > 
> > 	Rob Love
> > 
> > 
> 
> Perhaps synchronous was the wrong term =) But it does atleast seem to do 
> less buffering for removable devices or I could just be fooled by 
> something else slowing it down.

check under device manager and notice that there is an option checked on USB media disks that says "SYNC"?  I'm pretty sure that's pretty much synchronous transfers I assume.

Pat Erley

(Just delt with a neuros vs mvp3 chipset problem and noticed it)

Re: 2.7 (future kernel) wish

[Joshua Schmidlkofer]

> >
> Generally it just complains that you pulled out the device prematurely, 
> I've never seen one give a STOP error from that but I guess a bad driver 
> or USB controller could cause anything.
> 


Generally, yes.  But it is not over-statement to say 'sometimes'.  I
have seen this quite a bit. =(.  So I try not to pull devices w/o using
'Safely Remove...'

> When you insert a device like a USB stick Windows puts a little icon 
> next to the clock in the system tray that you're supposed to use to stop 
> the device before pulling it, effectively it unmounts and stops (or 
> atleast releases the device from) the driver so the device can be 
> 'safely' removed.

I am aware of this, =).  I was just commenting that yanking the device
in Windows - is not that much different from Linux.

Now, I would like to see the HAL type crap for Linux. It would be nice
to have uniform user device control from a gui programmable means.

js


-- 
VB programmers ask why no one takes them seriously, 
it's somewhat akin to a McDonalds manager asking employees 
why they don't take their 'career' seriously.

Re: 2.7 (future kernel) wish

[Elladan]

On Sat, Dec 27, 2003 at 10:03:17PM -0500, Jim Crilly wrote:
> >
> Generally it just complains that you pulled out the device prematurely, 
> I've never seen one give a STOP error from that but I guess a bad driver 
> or USB controller could cause anything.
> 
> When you insert a device like a USB stick Windows puts a little icon 
> next to the clock in the system tray that you're supposed to use to stop 
> the device before pulling it, effectively it unmounts and stops (or 
> atleast releases the device from) the driver so the device can be 
> 'safely' removed. I also believe Windows mounts any removable device 
> synchronously so that if you do pull it out prematurely the damage done 
> is limited.

I think the behavior of Win2k and WinXP is different for removable
devices.  Win2k seems to use async writes, and complains a lot if you
remove them without stopping.  XP seems to use sync writes, and doesn't
tend to complain much.

I haven't seen either one crash from pulling out flash cards from a card
reader (haven't tried a pen drive or similar).  However, the media
drivers they use for the USB readers sometimes seem to get into a bad
state where they're continually writing until you unplug the device.

-J

Re: 2.7 (future kernel) wish

[Helge Hafting]

On Sat, Dec 27, 2003 at 10:03:17PM -0500, Jim Crilly wrote:
> >
> >Sometimes Windows 2k or XP dump (BSOD), or maybe you just get an error. 
> >
> > 
> >
> Generally it just complains that you pulled out the device prematurely, 

Depends on what the device is used for, I guess.

> I've never seen one give a STOP error from that but I guess a bad driver 
> or USB controller could cause anything.
> 
Well, try having a partially loaded system dll on removable
media when you pull the plug - it won't be pretty.

> When you insert a device like a USB stick Windows puts a little icon 
> next to the clock in the system tray that you're supposed to use to stop 
> the device before pulling it, effectively it unmounts and stops (or 
> atleast releases the device from) the driver so the device can be 
> 'safely' removed. I also believe Windows mounts any removable device 
> synchronously so that if you do pull it out prematurely the damage done 
> is limited.

Linux has sync mounts too. :-)  the rest is a gui thing, i.e. not kernel.

Helge Hafting

Re: 2.7 (future kernel) wish

[Jim Crilly]

Helge Hafting wrote:
> On Sat, Dec 27, 2003 at 10:03:17PM -0500, Jim Crilly wrote:
> 
>>>Sometimes Windows 2k or XP dump (BSOD), or maybe you just get an error. 
>>>
>>>
>>>
>>
>>Generally it just complains that you pulled out the device prematurely, 
> 
> 
> Depends on what the device is used for, I guess.

Of course different things may happen depending on what's using the 
device, but IME I've never seen the OS give a STOP error from that.

> 
> 
>>I've never seen one give a STOP error from that but I guess a bad driver 
>>or USB controller could cause anything.
>>
> 
> Well, try having a partially loaded system dll on removable
> media when you pull the plug - it won't be pretty.

If you go through all the work to get Windows to use a system file from 
  removable device then pull it out while using it, you deserve what you 
get. The only thing you could do is page in the entire system file if 
you notice it's on a removable device then put a copy in swap for the 
case that someone pulls it out, but I don't think the special case would 
be worth it on Windows or Linux. Not that I have any proof Windows 
doesn't attempt to do that already =)

> 
> 
>>When you insert a device like a USB stick Windows puts a little icon 
>>next to the clock in the system tray that you're supposed to use to stop 
>>the device before pulling it, effectively it unmounts and stops (or 
>>atleast releases the device from) the driver so the device can be 
>>'safely' removed. I also believe Windows mounts any removable device 
>>synchronously so that if you do pull it out prematurely the damage done 
>>is limited.
> 
> 
> Linux has sync mounts too. :-)  the rest is a gui thing, i.e. not kernel.

I know, but I felt like mentioning the rest because it's relevant to a 
number of people on the list.

> 
> Helge Hafting

Jim.

RE: 2.7 (future kernel) wish

[Pacheco Jason NPRI]

> >>When you insert a device like a USB stick Windows puts a 
> >>little icon next to the clock in the system tray that you're 
> >>supposed to use to stop the device before pulling it, effectively 
> >>it unmounts and stops (or atleast releases the device from) 
> >>the driver so the device can be 'safely' removed.
> > 
> > 
> > This is useful, and something I think we need on the Linux 
> desktop (stay
> > tuned).
> > 
> 
> I agree, that's one of the reasons I posted at all. Little 
> things like 
> this can make a big difference, even though I've seen a lot 
> of users not 
> notice the little icon and have to be told about it.
> 
> Maybe when the icon appears have a tool-tip that pops up and says 
> something like "your USB device is ready for user at /mnt/usb, click 
> here when you're done" or something like that to make it more 
> noticable 
> that they shouldn't just yank it.
> 
> But I seem to be getting OT for this list...

Has anyone tried SuSE 9.0 (I believe the feature also exists in 8.2)?
It has a program called suseplugger that monitors hotswappable devices.
When I stick my USB key in it automatically detects and mounts the device
and presents an icon in KDE. It is very smooth and elegant, even more so
than in windows which pops up a bunch of "New Hardware Found" dialogs
each time you plug it into a different USB port for the first time. Also,
I unplugged the device a few times without unmounting it and there were
no complaints. In fact, I plugged it in again and was able to access the
files without having to remount it, Linux thought it was mounted the whole
time.

But I agree, OT, this is an interface issue.

-- P.S. Sorry Jim Crilly, set the To field wrong, my B

Re: 2.7 (future kernel) wish

[mjt]

On Tue, 30 Dec 2003 10:41:12 -0500,  Pacheco Jason NPRI <PachecoJ@Npt.NUWC.Navy.Mil> wrote:

> > But I seem to be getting OT for this list...
> 
> Has anyone tried SuSE 9.0 (I believe the feature also exists in 8.2)?
> It has a program called suseplugger that monitors hotswappable devices.
> When I stick my USB key in it automatically detects

... running suse 9.0 (and 8.2, and ...) and works as you describe
(say, with my usb camera). susewatcher/suseplugger existed as a
single module (app) in 8.x, but they've separated them into two
distinct apps in 9.0 

-- 
<<   http://michaeljtobler.homelinux.com/   >>
Important letters which contain no errors will develop 
errors in the mail.  Corresponding errors will show up 
in the duplicate while the Boss is reading it.