Skip to contentSkip to conversation

Release of 2.4.21

DISCUSSION Adrian Knoth 11 participants 17 messages Mar 20, 2003 – Mar 22, 2003

lore
Share & tools
lore
#0 lore raw

Hi,

how about releasing 2.4.21 with the ptrace()-fix applied immediately like it has been done with 2.2.25?

I think it's a serious bug and therefore it's time for a security-update.

If you think of ISC bind, PHP, sendmail or other software a new release follows right after the availability of the patch. I'd like Linux (and glibc as well) to do the same. You cannot call 2.4.20 a stable kernel with such a bug, so as a leader of the 2.4.x-series you cannot call the whole branch "stable".

World needs to update, best way to enforce is by a new release.

Signature · 3 lines
mail: [email protected]  	http://adi.thur.de	PGP: v2-key via keyserver

Das letzte Schoene, das in C geschrieben wurde, war Schuberts 9. Sinfonie.
#1 lore raw
Quoted 4 lines · Adrian Knoth

how about releasing 2.4.21 with the ptrace()-fix applied immediately like it has been done with 2.2.25?

I think it's a serious bug and therefore it's time for a security-update.

I think the best way is to release a 2.4.21 kernel with only the most important fixes (e.g. ptrace, ext3) and no new features. All new featues which need more testing and are now in 2.4.21-pre could then go to 2.4.22-pre for more testing (as Alan did with 2.2.25-pre1). This would be a way to react to important bugs very fast without a lack of enough testing which is just impossible with the current release scheme. It's just too dangerous to call 2.4.20 "stable" and wait another few month for 2.4.21.

Sebastian

#2 lore raw
Quoted 10 lines · Sebastian D.B. Krause
Quoted 4 lines · Adrian Knoth

how about releasing 2.4.21 with the ptrace()-fix applied immediately like it has been done with 2.2.25?

I think it's a serious bug and therefore it's time for a security-update.

I think the best way is to release a 2.4.21 kernel with only the most important fixes (e.g. ptrace, ext3) and no new features. All new featues which need more testing and are now in 2.4.21-pre could then go to 2.4.22-pre for more testing (as Alan did with

Something vaguely like this has been suggested, which I think is a good idea:

For critical fixes, release a 2.4.20.1, 2.4.20.2, etc. Don't disrupt the 2.4.21-pre cycle, that would be less productive than just patching 2.4.20 and rolling a separate release off of that.

Jeff

#3 lore raw
Quoted 3 lines · Jeff Garzik

For critical fixes, release a 2.4.20.1, 2.4.20.2, etc. Don't disrupt the 2.4.21-pre cycle, that would be less productive than just patching 2.4.20 and rolling a separate release off of that.

I think the naming is illogical. If there's a bugfix-only release it whould have normal incremental numbers. So if marcelo want's it he should clone a tree of at 2.4.20, apply the essential patches and bump the version number in the normal 2.4 tree to 2.4.22-pre1

#4 lore raw
Quoted 6 lines · Christoph Hellwig
Quoted 3 lines · Jeff Garzik

For critical fixes, release a 2.4.20.1, 2.4.20.2, etc. Don't disrupt the 2.4.21-pre cycle, that would be less productive than just patching 2.4.20 and rolling a separate release off of that.

I think the naming is illogical. If there's a bugfix-only release

Many, many companies seem to find it logical. If you want to squeeze a version in between "1" and "2".

Further, other kernel hackers suggested the 2.4.20.N sequence, I simply agreed with it. So it's not only me who thinks this way :)

Quoted 3 lines

it whould have normal incremental numbers. So if marcelo want's it he should clone a tree of at 2.4.20, apply the essential patches and bump the version number in the normal 2.4 tree to 2.4.22-pre1

Human nature says that will drag out the -pre tree ad infinitum. Suppose a 2.4.21 is released today, with 2.4.20 + bug fixes. Now, tomorrow, another "critical bug" comes out, and then the -pre tree becomes 2.4.23-pre. Add another critical bug, and I hope you see the continual delay of -pre happens here...

The basic logic is "do not disrupt current plans. Do something _in addition to_ current plans."

Jeff

#5 lore raw

hopefully we won't have a steady stream of critical bugs. if we do we need to stop new development and find out where we are going wrong.

after all this isn't microsoft.

one advantage of useing a good version control system is that it should be very easy to make a minor update to 2.4.20 to produce 2.4.21 without it being a major disruption to things and then continue with the existing development. from what I understand about bk it should make it even easier.

I don't like going to 4 part version numbers. if this was a marketing driven company that had promised features in 2.4.21 and they aren't ready but a new version is needed then we would have a reason to do a 2.4.20.1 so that we don't have to deal with broken promises, but sine we aren't in that situation I don't see why the 2.4.20.N is needed.

Every other time there has been a critical bug it was fixed in the next normal release number (or if it's critical enough it was handled by rushing a version with just that fix like happened with 2.2)

David Lang

Quoted 44 lines · Jeff Garzik

Date: Thu, 20 Mar 2003 15:53:38 -0500
From: Jeff Garzik <[email protected]>
To: Christoph Hellwig <[email protected]>, [email protected],
[email protected]
Subject: Re: Release of 2.4.21

Quoted 6 lines · Christoph Hellwig
Quoted 3 lines · Jeff Garzik

For critical fixes, release a 2.4.20.1, 2.4.20.2, etc. Don't disrupt the 2.4.21-pre cycle, that would be less productive than just patching 2.4.20 and rolling a separate release off of that.

I think the naming is illogical. If there's a bugfix-only release

Many, many companies seem to find it logical. If you want to squeeze a version in between "1" and "2".

Further, other kernel hackers suggested the 2.4.20.N sequence, I simply agreed with it. So it's not only me who thinks this way :)

Quoted 3 lines

it whould have normal incremental numbers. So if marcelo want's it he should clone a tree of at 2.4.20, apply the essential patches and bump the version number in the normal 2.4 tree to 2.4.22-pre1

Human nature says that will drag out the -pre tree ad infinitum. Suppose a 2.4.21 is released today, with 2.4.20 + bug fixes. Now, tomorrow, another "critical bug" comes out, and then the -pre tree becomes 2.4.23-pre. Add another critical bug, and I hope you see the continual delay of -pre happens here...

The basic logic is "do not disrupt current plans. Do something _in addition to_ current plans."

Jeff

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

#6 lore raw
In reply to Christoph Hellwig
Quoted 10 lines · Christoph Hellwig
Quoted 3 lines · Jeff Garzik

For critical fixes, release a 2.4.20.1, 2.4.20.2, etc. Don't disrupt the 2.4.21-pre cycle, that would be less productive than just patching 2.4.20 and rolling a separate release off of that.

I think the naming is illogical. If there's a bugfix-only release it whould have normal incremental numbers. So if marcelo want's it he should clone a tree of at 2.4.20, apply the essential patches and bump the version number in the normal 2.4 tree to 2.4.22-pre1

No point in making things too complex. 2.4.20-post1 is something people can easily understand.

I needed that for the ext3 problems which popped up shortly after 2.4.20 was released - I was reduced to asking people to download fixes from my web page.

And having a -post stream may allow us to be a bit more adventurous in the -pre stream.

#7 lore raw
Quoted 9 lines
Quoted 8 lines
Quoted 3 lines

For critical fixes, release a 2.4.20.1, 2.4.20.2, etc. Don't disrupt the 2.4.21-pre cycle, that would be less productive than just patching 2.4.20 and rolling a separate release off of that.

I think the naming is illogical. If there's a bugfix-only release it whould have normal incremental numbers. So if marcelo want's it he should clone a tree of at 2.4.20, apply the essential patches and bump the version number in the normal 2.4 tree to 2.4.22-pre1

Quoted 8 lines

No point in making things too complex. 2.4.20-post1 is something people can easily understand.

I needed that for the ext3 problems which popped up shortly after 2.4.20 was released - I was reduced to asking people to download fixes from my web page.

And having a -post stream may allow us to be a bit more adventurous in the -pre stream.

Why can't we just make all releases smaller and more frequent?

Why do we need 2.4.x-pre at all, anyway - why can't we just test things in the -[a-z][a-z] trees, and _start_ with -rc1?

Why can't we just do bugfixes for 2.4, and speed up 2.5 development?

John.

#8 lore raw
Quoted 22 lines · John Bradford
Quoted 11 lines
Quoted 11 lines
Quoted 5 lines

For critical fixes, release a 2.4.20.1, 2.4.20.2, etc. Don't disrupt the 2.4.21-pre cycle, that would be less productive than just patching 2.4.20 and rolling a separate release off of that.

I think the naming is illogical. If there's a bugfix-only release it whould have normal incremental numbers. So if marcelo want's it he should clone a tree of at 2.4.20, apply the essential patches and bump the version number in the normal 2.4 tree to 2.4.22-pre1

Quoted 10 lines

No point in making things too complex. 2.4.20-post1 is something people can easily understand.

I needed that for the ext3 problems which popped up shortly after 2.4.20 was released - I was reduced to asking people to download fixes from my web page.

And having a -post stream may allow us to be a bit more adventurous in the -pre stream.

Quoted 9 lines

Why can't we just make all releases smaller and more frequent?

Why do we need 2.4.x-pre at all, anyway - why can't we just test things in the -[a-z][a-z] trees, and _start_ with -rc1?

Why can't we just do bugfixes for 2.4, and speed up 2.5 development?

That would imply some changes could take place in a short cycle. This
is not true for things like major ide subsystem updates.

Signature · 4 lines
There is no such thing as obsolete hardware.
Merely hardware that other people don't want.
(The Second Rule of Hardware Acquisition)
Sam Flory  <[email protected]>
#9 lore raw
Quoted 27 lines
Quoted 27 lines
Quoted 11 lines
Quoted 11 lines
Quoted 5 lines

For critical fixes, release a 2.4.20.1, 2.4.20.2, etc. Don't disrupt the 2.4.21-pre cycle, that would be less productive than just patching 2.4.20 and rolling a separate release off of that.

I think the naming is illogical. If there's a bugfix-only release it whould have normal incremental numbers. So if marcelo want's it he should clone a tree of at 2.4.20, apply the essential patches and bump the version number in the normal 2.4 tree to 2.4.22-pre1

Quoted 10 lines

No point in making things too complex. 2.4.20-post1 is something people can easily understand.

I needed that for the ext3 problems which popped up shortly after 2.4.20 was released - I was reduced to asking people to download fixes from my web page.

And having a -post stream may allow us to be a bit more adventurous in the -pre stream.

Why can't we just make all releases smaller and more frequent?

Why do we need 2.4.x-pre at all, anyway - why can't we just test things in the -[a-z][a-z] trees, and _start_ with -rc1?

Quoted 7 lines
Quoted 4 lines

Why can't we just do bugfixes for 2.4, and speed up 2.5 development?

That would imply some changes could take place in a short cycle. This
is not true for things like major ide subsystem updates.

We should not have major updates like that as a regular occurance in the stable kernel series. If they are necessary, we could break with the small, frequent updates, and go back to what we have now.

Why can't we basically have two modes of working:

* When nothing major is being re-written:

Small, frequent updates. -pre and -rc version numbering used to indicate kernels which haven't had much testing. Larger updates existing in -[a-z][a-z] trees.

* When an update to a core subsystem is in progress

Larger, less frequent updates. -pre and -rc versions getting more testing, and -[a-z][a-z] trees for very experimental code.

I know there are frequently suggestions about how to organise kernel development, and they are usually ignored, because we generally don't like to work however we want to, but this suggestion is just common sense. When nothing much is going on, release smaller changes to get a lot of testing of the same codebase. When major work is going on, separate out the minor updates to give people flexibility in testing.

John.

#10 lore raw
In reply to John Bradford
Quoted 6 lines · John Bradford
Quoted 1 line

Why can't we just make all releases smaller and more frequent?

Why do we need 2.4.x-pre at all, anyway - why can't we just test things in the -[a-z][a-z] trees, and _start_ with -rc1?

Why can't we just do bugfixes for 2.4, and speed up 2.5 development?

Because 2.4 is used on (real) production systems - even my desktop PC on my workplace is considered a production system - which (at least) should run and therefore trying 2.5 is not an option (and no, no way). And then it takes 1.5 years for the next stable kernel release which implies that quite new hardware (without an existing driver) cannot be used for any production-like system.
IOW you would just loose a lot real use and testing of backported stuff and new hardware drivers.
And no, I don't think that someone wants that.

Bernd

Signature · 4 lines
Bernd Petrovitsch                              Email : [email protected]
g.a.m.s gmbh                                  Fax : +43 1 205255-900
Prinz-Eugen-Straße 8                    A-1040 Vienna/Austria/Europe
                     LUGA : http://www.luga.at
#11 lore raw
Quoted 5 lines
Quoted 5 lines
Quoted 1 line

Why can't we just make all releases smaller and more frequent?

Why do we need 2.4.x-pre at all, anyway - why can't we just test things in the -[a-z][a-z] trees, and _start_ with -rc1?

Quoted 8 lines
Quoted 1 line

Why can't we just do bugfixes for 2.4, and speed up 2.5 development?

Because 2.4 is used on (real) production systems - even my desktop PC on my workplace is considered a production system - which (at least) should run and therefore trying 2.5 is not an option (and no, no way). And then it takes 1.5 years for the next stable kernel release which implies that quite new hardware (without an existing driver) cannot be used for any production-like system.

I would include support for new chipsets in with bugfixes.

Quoted 2 lines

IOW you would just loose a lot real use and testing of backported stuff and new hardware drivers.

If nobody tests 2.5.x on a specific piece of hardware, 2.6.0 will be released without being tested on that piece of hardware.

Incrementing the version number from 2.5.$bignum to 2.6.0 does _not_ magically fix all the bugs in it.

Quoted 1 line

And no, I don't think that someone wants that.

Production systems should be backed up. No question about that.

Unless the system has to run 24/7, which a desktop machine usually doesn't have to, I don't see why testing 2.5 on it isn't a possibility.

John.

#12 lore raw

Am Fre, 2003-03-21 um 10.23 schrieb John Bradford:

Quoted 3 lines

Unless the system has to run 24/7, which a desktop machine usually doesn't have to, I don't see why testing 2.5 on it isn't a possibility.

Because desktop systems usually don't feature the notion of a testbed, read an environment in which dangerous changes can be tested without severe loss of important data.

I would never ever boot a 2.5.x kernel on my notebook or on any other machines without preparing a different harddrive for instance, YMMV though.

Signature · 2 lines
Servus,
       Daniel
#13 lore raw
Quoted 7 lines
Quoted 3 lines

Unless the system has to run 24/7, which a desktop machine usually doesn't have to, I don't see why testing 2.5 on it isn't a possibility.

Because desktop systems usually don't feature the notion of a testbed, read an environment in which dangerous changes can be tested without severe loss of important data.

So why did you trim the part of my quote that said that production systems should always be backed up anyway?

Quoted 3 lines

I would never ever boot a 2.5.x kernel on my notebook or on any other=20 machines without preparing a different harddrive for instance, YMMV though.

It's not exactly much work to put a different disk in a machine. With a notebook, it might be 30-45 minutes work dismantling it, but on a typical desktop, about 2 minutes, and on a rackmount server, probably 60 seconds.

John.

#14 lore raw

Am Sam, 2003-03-22 um 09.27 schrieb John Bradford:

Quoted 2 lines

So why did you trim the part of my quote that said that production systems should always be backed up anyway?

You consider normal desktop systems "production system"? Interesting... The best on can hope for a desktop system is a regular becakup of important work files to CD-R; we're more talking about a sector-wise harddisk backup here which is not even common for server systems (no, I'm not talking about RAID here) let alone Joe users system.

For instance I do have a RAID and a regular rsync backup on my production system here. That's still not enough for trying development kernels on real data which might fry the filesystem because it'd be fried on the mirrored harddrives at the same time and the rsynced copy is not enough to fully recover the system.

Quoted 1 line

It's not exactly much work to put a different disk in a machine.

Yeah, though not something Joe user could do. And honestly this is not something someone would do "just for fun", because it's still a lot of trouble for a minor intent.

Quoted 1 line

With a notebook, it might be 30-45 minutes work dismantling it,

You certainly must be kidding. Dismantling my PowerBook is hardly possible at all without killing it and the harddrive is quite hard to replace. This is more like a 2h work and something I'd rather do once in its lifetime for the sake of a bigger harddrive instead of trying a development kernel.

Quoted 2 lines

but on a typical desktop, about 2 minutes, and on a rackmount server, probably
60 seconds.

Right. Let's see. It took me about 20 minutes to swap the CPU in this "desktop" machine here, 15 minutes of that were used to dismantle the system, pull the mounted motherboard (with all PCI cards in the slots, my case luckily allows for that) and reassemble those parts. This action is necessary to get to the right hand side screws which fasten the harddrive.

It's not even realistic to assume that simply opening the case, and reconnecting a loosely hanging around harddrive will just take 2 minutes.

Signature · 2 lines
Servus,
       Daniel
#15 lore raw
In reply to Adrian Knoth
Quoted 5 lines · Adrian Knoth

(and glibc as well) to do the same. You cannot call 2.4.20 a stable kernel with such a bug, so as a leader of the 2.4.x-series you cannot call the whole branch "stable".

World needs to update, best way to enforce is by a new release.

The vendors have released updated kernels. Whats the problem ?

#16 lore raw

if official kernels were being released more rapidly then they are it wouldn't be much of a problem (I can wait a week or two for this with no problem) but it's currently several months between releases, and since there are so many changes there is the potential that a new kernel could be worse then what you are running.

yes I could download a patch and fix the kernel that way, but while I understand how simple that is I can already hear the crys from management about haphazardly patching things that then get put into production.

and as for the vendor kernels, I don't use them becouse they all make a ton of changes that I don't want to deal with.

the stable kernel releases are a vendor for this situation and this 'vendor' has chosen to consider this bug not critical (along with a couple others) which is a decision being questioned by folks.

David Lang

Quoted 21 lines · Alan Cox

Date: 21 Mar 2003 01:01:34 +0000
From: Alan Cox <[email protected]>
To: Adrian Knoth <[email protected]>
Cc: Linux Kernel Mailing List <[email protected]>
Subject: Re: Release of 2.4.21

Quoted 5 lines · Adrian Knoth

(and glibc as well) to do the same. You cannot call 2.4.20 a stable kernel with such a bug, so as a leader of the 2.4.x-series you cannot call the whole branch "stable".

World needs to update, best way to enforce is by a new release.

The vendors have released updated kernels. Whats the problem ?

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