Submitted by Linux 2.6.15 has been released after two months and a week of development. You can check the comprehensible changelog or grep the full changelog. There are some new features like shared subtrees, UDP fragmentation offload, PPP MPPE encryption (VPN), NTFS write support (except for creating files), PPC64 thermal improvements, support for the late-2005 powerbook series, SATA passthru support (for SMART), console rotation for fbcon, nf_conntrack subsystem, some scalability and performance improvements, and lots of other changes. As always, download it from Kernel.org, or wait for your distributor to ship it.
There are a lot of good improvements here. I am particularly interested in the NTFS “write” and the console rotation. The shared subtrees looks interesting, but I need to read into it a little more.
i dont get it. ntfs write support has been in the kernel for a long time.. altough always marked as experimental and dangerous.
does this means that it is now safe to use?
If you search for ntfs in the full changelog, there are quite a lot of new patches. This new work seems interesting, has anyone used it yet that can post some feedback from the -rc’s?
With the defaults, writing to an NTFS partition with an OS other than that which created the partition is always a risk proposition. NTFS by default, i.e. with a “basic” disk, relies too heavily on the registry, so you can corrupt data by writing to it when you are not using the same registry. I presume this is why it currently says you cannot create new files. Using dynamic disks the data is instead stored in the partition itself, making the partition much more portable, but I don’t know how well NTFS dynamic disks are supported outside of Windows.
Its a mess.
Damien
NTFS by default, i.e. with a “basic” disk, relies too heavily on the registry, so you can corrupt data by writing to it when you are not using the same registry.
Nonsense. Source please.
Well, one can always use captive-ntfs, utilizing Windows DLLs. That is, if it works. For me it works occasionally but it fails more often than not.
Another option is to install the freeware ext2fs driver for Windows and use ext2fs for shared partitions.
(NOTE: It mounts ext3fs as ext2fs.)
The best option for me is a 2nd PC with CIFS/Samba. I use my “closet server” (it’s a server tower with 3 30g SCSI disks sitting in my closet…doing random odd network jobs, web server, mysql, ftp, file storage etc.) to copy files too and fro over gigabit. It’s not entirely fast sometimes but it works quite well most of the time.
ext3ifs works great, too bad for those of us with reiserfs
The reiserfs drivers hasn’t worked well for me.
> NTFS by default, i.e. with a “basic” disk, relies too
> heavily on the registry, so you can corrupt data by
> writing to it when you are not using the same registry.
Does the registry just contain the according policy settings, or does it contain nessecary information to read/write the disk? Some links on this topic would be nice. Especially, what problems could appear when using an NTFS disk (‘secondary’, without access to the registry of the creating OS) in another NT-based OS?
– Morin
I have never heard of the registry required for writes. Some storage services tie to the registry like quotas lanman shares. Being a SAN guy I move NTFS volumes around all the time!!!!
altough always marked as experimental and dangerous
So far, there have been two NTFS write methods in Linux. You are talking about the old one, which was indeed dangerous. But for some time, it has been replaced by a non-dangerous method. The drawback is that it only allows to modify the content of an existing file without changing its size. This is useful, for instance, for Topologilinux, the distro that can be installed inside a windows partition.
Apparently, these limitations have been softened, which is good.
“The drawback is that it only allows to modify the content of an existing file without changing its size.”
Are you sure this is still the case? I can’t see why they would even bother including the ability to write to NTFS if the file sizes must remain the same.
That is almost like saying you can make changes as long as the md5 hash of the completed work ends up the same.
Edit, text removed. Buffer added to exactly match file size of previous comment. XXXXXXXXXXXXXXXXXXXXX
Edited 2006-01-03 18:30
Are you sure this is still the case? I can’t see why they would even bother including the ability to write to NTFS if the file sizes must remain the same.
In fact, it seems most writing is safe now. From the changelog:
commit 98b270362bb9ea6629732e7f5b65b8a6ce4743c7
Author: Anton Altaparmakov <[email protected]>
Date: Tue Oct 11 15:40:40 2005 +0100
NTFS: The big ntfs write(2) rewrite has arrived. We now implement our own
file operations ->write(), ->aio_write(), and ->writev() for regular
files. This replaces the old use of generic_file_write(), et al and
the address space operations ->prepare_write and ->commit_write.
This means that both sparse and non-sparse (unencrypted and
uncompressed) files can now be extended using the normal write(2)
code path. There are two limitations at present and these are that
we never create sparse files and that we only have limited support
for highly fragmented files, i.e. ones whose data attribute is split
across multiple extents. When such a case is encountered,
EOPNOTSUPP is returned.
I wish they’d sort it out so I could do thinks like edit ID3 tags when I’m mounting my NTFS drives under Linux (I dual boot, spend most of my time in Linux but keep MP3s o NTFS so I can access them when I’m in either OS). It would be useful.
But anyway, Linux keeps getting better…
Why not use FAT32. Simpler, more OS-es know it, linux has full write access. I use it for same purpose you do.
Files <= 4GB really suck when you like to rip dvds on windows and encode them on linux clusters. Fat32 has a 4GB filesize limitation. If you have a ton of 4GB files, it tends to get very fragmented and slow the system down unless you defrag once a week.
Now I just rip + encode from linux and don’t worry about that anymore.
FAT32 is extremely unreliable, very slow and fragments extremely fast.
Why not use ext3fs instead? Install the freeware Windows ext2fs driver (works with ext3fs as well – mounted without journalizing though), and off you go. Full read/write support in windows AND linux. Much much better
Still no Reiser4
The Same Here….
Still No Reiserf4 🙁
You can always patch the kernel yourself, and recompile it. The problem is that there are only two ways Reiser 4 will ship:
1) Every file-system in the kernel is upgraded to support Reiser 4 functionality, even at a basic “pretend-to-work-but-do-nothing” level
2) Every user-space tool is rewritten to support both vanilla file-systems and Reiser 4 file-systems.
Neither is trivial, and currently the kernel developers are unwilling to do (1) as the Reiser people are unwilling to help them with it.
It’s a bit of a sticky mess, and it would really help if there was a genuine development tree, but there you go. Hopefully some brave individual (or distributor) will see some sense and do the initial work to get kernel developers interested in attempting this.
Is it even close to being stable? The last time I tried it, 1 or 2 months ago the maskine keept freezing even the reiser4-partition had just a little load.
I havent upgraded to the latest and greatest kernel still at 2.4…but uhh…is it just me or are they trying to cram everything into the kernel? Is that a good idea? Are there benefits from having VPN and support for PPC in kernels that will be running on say x86 procs?
If you’re running on x86 the PPC support won’t have got built at all, so the only effect on you is that the source code is a bit bigger. The VPN support may be built, if you (or your distributor) configure it but even then, with a modular kernel the extra code won’t get loaded into memory unless it’s actually used.
If that was you who voted that person down, please don’t vote someone down because they’re ill-informed.
I take you haven’t ever built a kernel yourself? Try gentoo stage 1 sometime, it’s very enlightening, if nothing else one can really learn a lot about the basics of a linux system just by installing and using gentoo for a few months.
In fact, almost any source based distro is that way.
The linux kernel is extremely scaleble and most distro creators only build what they think is absolutely needed into the kernel, everything else is built as modules that are inserted into the kernel whenever they’re needed, either by the user for by a subsystem.
Depending on the distro the majority of non-desktop oriented kernel modules are never even built. And things like PPC stuff aren’t even ever considered or turned on when the target arch is anything but PPC.
Thanks. I am not used to compiling the kernel. I have been thinking about doing a Gentoo stage 1 install actually but I need a new machine…or an old…cause right now all my machines are Windows based because of the DCC work I am doing. Still I was curious is all. If I got modded down that is fine.
Full NTFS write support will be one of those “killer features” of the Linux desktop allowing many people to try out a dual boot who might have been hesitant before.
Maybe the broadcom drivers from the BCM43xx project will be integrated into 2.6.16. Allowing full working wireless on ibooks and most of the HP laptops (zxxx series for sure). Damn I love open source
Would be better to support the ext2fs driver for windows
I disagree… We are trying to allow users to migrate *to* linux easier, not back to windows
Yup. And I believe the best way to do so, is by replacing non-FLOSS technologies on the windows platform with FLOSS technologies, rather than implementing non-FLOSS technologies on FLOSS systems.
>Why not use FAT32. Simpler, more OS-es know it,
>linux has full write access. I use it for same
>purpose you do.
FAT32 is really slow. But you’re right, for what I use it for, it would make sense.
http://www.google.com/search?q=cache:IT1X2f-H0yUJ:wiki.kernelnewbie…
Luckily I have no 4GB files. All my MP3s are from my own CDs, too. Not a ‘net leech.
http://www.google.com/search?q=cache:IT1X2f-H0yUJ:wiki.kernelnewbie…
This is a honest question from someone who does not know truly the mechanics of a filesystem. What about making a “portable” filesystem? What I mean by this is, we have reiserFS just sitting there. Would not it be possible to have someone (now, the question is who can handle that herculean task) figure a way to make the said filesystem run perfectly fine and natively (well it’s still a hack, but you get the idea) under Microsoft Windows, and maybe Mac OS X?
What are the odds of that happenning?
I remember a project of “reiser driver” that used some development kit, however, I have never been able to make that work, I have not seeked further to run it properly, I think it was beyond my scope and it did not “behave” like a filesystem, meaning you would not be able to use it from “my computer” and such. That defeats my purposes somewhat.
http://rfsd.sourceforge.net/
Anyway, I’m just curious. I’m windowless since exactly one year and I’m not ever going back, however, I would be interested to see an “open” filesystem that can cross those boundaries. It would greatly ease the conversion of my friends.
This is a honest question from someone who does not know truly the mechanics of a filesystem. What about making a “portable” filesystem
There IS a portable filesystem, supported by approx 80% of general purpose operating systems. It’s called FAT32. Linux supports it, all versions of Windows 95 and up support it, Mac OS 9 and OS X support it, Bluebottle does, Plan 9 does, BeOS does, AROS does, the BSDs do, probably RISC OS and AmigaOS as well, as well as a whole slew of other OSes.
The problem is that FAT32 isn’t very advanced and has it’s limits, but if you want something advanced, you lose the portable aspect very quickly.
One step up from FAT32 would perhaps be EXT2, which is supported on BeOS, Windows (according to post above), Mac, BSDs and of course Linux.
Edited 2006-01-03 20:25
Yeah, sorry I forgot to add the mention I was looking for a “modern” alternative
keeping file permissions, being FAST, decent filename/volname length and so on… I’m not running a nuclear plant cluster, I’m just a “normal” user, I’ve come to “expect” those features from a filesystem though.
I know …
I can dream can I?
I guess most Linux supported filesystems are portable in a sense. Their on-disk structures are documented and the algorithms they use are known, sou you could implement a driver for any one of them in another OS.
The real problem is the complexity. The complex ones, with good fetures and performance need a great deal of code to drive them, and more code means more dependencies on the kernel they were originally designed on/to. It is still possible to port them to another OS, but it means more work (e.g. XFS was ported to Linux from IRIX, I guess most of the code is new in Linux, but it can work with partitions created on IRIX).
That said, I remember reading somewhere how the cost and licensing for the Windows Filesystem SDK was the real impediment for cross-platform(OS) filesystem development on windows.
Anyone seen any problems with 2.6.15 with NetworkManager…running 3.4 from fc3 on centos4.2 and its dying for no apparent reason. Still investigating..
Linux is light years from FreeBDSD 6 in terms of performance.
Based on what? do you have any numbers? or just a feeling?! If you dont have any stats/bench or numbers you should improve your speech!
Ahhh, the obligatory worthless comment.
If gnome is posted then some KDE nerd will post something about KDE being better. If KDE is posted then some gnome nerd will post about gnome being better. And aparently when Linux is posted some FreeBSD nerd will post about it being better.
Nerds take things too personally.
“Nerds take things too personally.”
Are you saying I’m a nerd? Wtf, just because I have no life, I’m not a nerd! Totally uncalled for. Hey, and Gnome _is_ better than KDE. KDE suxxorz and Linux rulez!
Edited 2006-01-04 13:37
Are they ever going to stabiliz the 2.6 kernal and start a 2.7 dev kernal. I don’t really like the fact that features are being added to the main branch of the kernel…I like stability, and I don’t think i am the only one
Features are not orthogonal to stability. As long the changes you make are not intrusive like adding a a new driver which doesnt require changing existing infrastructure or if you are fixing things which are already known to be broken like better write support for a filesystem then there isnt a problem with changing stuff.
Unfortunately the changes are all relatively intrusive, with many significant API changes, thus stability is comprised (which is why Greg KH is working on his 2.6.x.y semi-stable tree). What’s more highly intrusive work with the potential for a big payoff (real-time Linux, Reiser4 integration / fs upgrades) is rejected out of hand as it would be too hard to get any sort of a release out the door in a reasonable amount of time.
The result is something that’s fun to work on, but hard for distributors to maintain (I dread to think how the Debian developers are maintaining 2.6.8 for example, given that it’s totally orphaned at this stage), and which also sees no major work in hard areas.
“Unfortunately the changes are all relatively intrusive, with many significant API changes, thus stability is comprised (which is why Greg KH is working on his 2.6.x.y semi-stable tree). ”
-mm tree is the unstable tree now where changes are put before getting into mainline
Majority of API changes are internal and do not affect user space utilities and that was the case in Linux 2.4 series too. Kernel developers dont care about third party non upstreamed patches and modules which are the only ones getting affected by such API changes
Linux 2.6.x.y series contains critical security and bug fixes which are released until the subsequent 2.6.x release to avoid the time delay between releases affecting users.
No they are not. Having one branch has the advantage there’s no need to backport features to old kernel. Also new features come available sooner.
Actually there are two ‘branches’, 2.6 and -mm which contains experimental patches. When the code in -mm has been proven stable and useful enough it is merged to mainline.
Stability might suffer a little though but not much really because there’s two weeks time frame after a new release (like 2.6.15) when new features are added. After that only bug fixes are accepted. Release candidates are released until kernel becomes stable. As you can see from the changelog there are only small changes after the latest -rc.
Of course 2.6.15 can still contain some bugs (especially in drivers) if not enough people have tested the -rc kernels.
Are you personally experiencing stability issues related to the kernel? My issues tend to revolve around X server and higher up the stack personally (and even those are not common).
“Hey, it’s fifteen years today since I bought the machine that got Linux started. January 2nd is a good date.” –Linus Torvalds
You can learn quite a bit by just compiling a kernel, playing with the config system, looking at what’s produced, etc. Also, if you don’t want to mess with your main installation, you may be able to build a User Mode Linux kernel yourself, then test it in a UML virtual machine. This means you don’t have to mess with your existing setup – you should be able to find a howto on the web. (btw, I didn’t mod you down – twas a valid question to ask)
Does this mean anyone with a Knoppix CD can bypass the standard file protection scheme on Windows machines?
anybody with a windows live cd can get past ntfs permissions. unless it is encrypted ntfs filesystem then you need the correct user and password to decrypt
Does this mean anyone with a Knoppix CD can bypass the standard file protection scheme on Windows machines?
Yep, but of course that same Knoppix CD can bypass the file permissions on any Linux partitions too.
If someone has physical access to a machine only encryption can really protect your data.
Locking the BIOS to stop it from booting from CD might deter an attacker though, as he’ll now have to open the machine to zap the CMOS or take out the hard drive. Put a mechanical lock on it as well, and he’ll have to bring a crowbar or take the whole machine to break into it later.
File permissions only work for regulating remote access and for preventing accidental damage.
I hope that SATA stability is improved on the amd64 platform with nforce4! At high loads (unpacking some file while watching TV for instance) the sata driver will crash, making the disk inaccessable. Eventially the system will just crash
I also hope that the SMART stuff finally works.
Yeah those stupid kernel developers keep adding features for some reason!! But seriously, since the new development model took over, I haven’t had any stupid things happen like I did from 2.6.8 – 2.6.12 it’s been fairly solid since then for me. If you’re running a desktop then I’d say you really want to run 2.6 if you’re running a server, then it might not be as obvious, but it is likely that you’d get better performance from a more modern kernel.
…UDP fragmentation offload…
Will this improve things like BitTorrent (DHT creates massive amounts of UDP packets for instance), or is this something else entirely? Either way, it looks like it’s for a good cause. 🙂
It seems that the stability has been indeed highly improved. I’ve been testing it yesterday (and today morning) intensively and until now I see no issues with my SATA-II drive.
Tested on A8N-VM CSM mainboard (with manually patched DSDT ACPI table – the original one is broken)