Announcement

Collapse
No announcement yet.

Security Problem Discovered In Btrfs File-System

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Originally posted by crazycheese View Post
    First - ZFS is Solaris exclusive and only ported to BSD.
    Solaris has the best implementation and FreeBSD has the best port, but ZFS is by no means Solaris-exclusive. There are dozens of operating systems that support it if you consider each distribution to be a different OS. In specific, the following major operating system families have some level of ZFS support:

    Darwin
    FreeBSD
    Linux
    NetBSD
    Solaris/Illumos
    Windows NT

    Originally posted by crazycheese View Post
    Second - ZFS inferior to BTRFS in many operations.
    Would you name some?

    Originally posted by crazycheese View Post
    Third - ZFS is different and for different scale, many complexities are excessive for different systems. ZFS is meant for datacenters.
    The same could be said for Linux, yet people who use Linux on desktops would disagree. Similarly, people who use ZFS on desktops (such as myself) would disagree. Having used ZFS on my desktop, I consider it to be the best filesystem available for systems that run mainstream Linux distributions and I strongly encourage its use.

    Originally posted by crazycheese View Post
    Six - ZFS also has limitations.
    Do these limitations include Denial of Service? I have managed to hang systems that use a combination of ext4, CFQ and discard. I have yet to hang a system using ZFS outside of situations that involve experimental patches.

    Originally posted by crazycheese View Post
    Seven - ZFS developers very very rarely accept patches to improve its "desktop" usage. See (3).
    Would you elaborate on that? I have written dozens of patches to improve the ZFS Linux port that resulted in changes to the upstream repository. All of them were developed on my desktop.

    Originally posted by crazycheese View Post
    The only fact is that ZFS is purposely not compatible to GPL.
    What is your point? The GPL is purposely incompatible with a wide range of licenses. In the case of the CDDL, the incompatibility only affects distribution of a kernel binary containing ZFS.

    Originally posted by pankkake View Post
    It is very true that ZFS isn't very good for desktop usage.
    Would you elaborate? People usually tell me that using ZFS as their rootfs makes their computers perform faster. That has been my experience.

    Originally posted by pankkake View Post
    btrfs is much more universal.
    How can btrfs be more universal than ZFS when it is limited to Linux?
    Last edited by ryao; 12-15-2012, 03:10 PM.

    Comment


    • #17
      Originally posted by crazycheese View Post
      Was that hard to even use Wikipedia properly?
      "ZFS was designed and implemented by a team at Sun led by Jeff Bonwick and Matthew Ahrens. It was announced on September 14, 2004,[5] but development started in 2001.[6] Source code for ZFS was integrated into the main trunk of Solaris development on October 31, 2005[7] and released as part of build 27 of OpenSolaris on November 16, 2005. Sun announced that ZFS was included in the 6/06 update to Solaris 10 in June 2006, one year after the opening of the OpenSolaris community."

      Is this hard to understand "ZFS was Solaris exclusive"? Can you distinguish "original platform" and "port platform"? I am sure you can.
      Sun made it. Sun was author of Solaris. This is very illogical,no?
      Then make up your mind with this statement:

      Originally posted by crazycheese View Post
      First - ZFS is Solaris exclusive and only ported to BSD.
      Note: I added the bold/underline

      ZFS was conceived in Solaris (d'oh !), but not exclusive of this platform ... even if you "only ported to BSD" is still wrong, you say the opposite you said before with the "exclusive".

      I SEE WHAT YOU DID THERE

      First of all, Wikipedia is outdated in some areas regarding ZFS.
      Second → I'm sorry, but BSD is not an OS (I'll take it as a OS family) ... and again

      With only this, I invalidate your point → http://zfsonlinux.org/
      With this too → http://code.google.com/p/maczfs/
      Again, with this too → https://www.haiku-os.org/tags/zfs
      Also this → https://duckduckgo.com/?q=IllumOS

      Not counting the different *BSD systems.
      I prefer to talk about the community and free ZFS, not the Oracle's one ... ty very much

      Second - ZFS inferior to BTRFS in many operations. Many times it looses because its just too complex. Other times it looses due to design. It is more polished, but it is different. Compare FAT32 with EXT4 in data ordered mode - you get equal numbers, EXT4 will loose. Is this bad? No.
      First of all, citing a benchmark from Michael Larabel is something I always "take with tweezers"
      Michael is well known for his "well" and badly done benchmarks mostly because he had no formation on what he was benchmarking ... I remember a really horribly made benchmark (in fact, if my memory doesn't fail in one of the "benchmarks" he didn't use the same hardware) that ended up in a flame war in the mailing lists and this wiki page was born from that: http://wiki.freebsd.org/BenchmarkAdvice

      That article FROM 2010 lacks information about OS, configuration, specific filesystem configurations and some extra information to actually extract something meaninful from these numbers ... sorry, can't take that seriously, but those are cute colored graphs nonetheless
      When it comes to linux graphical stack benchs, some people here also remembers horribly done benchmarks ...

      Do you have anything with a little bit more substance?

      Third - ZFS is different and for different scale, many complexities are excessive for different systems. ZFS is meant for datacenters. You want to use ZFS only if you fear bit-rot, but the performance will be abysmal and most features will simply be outside of scope of desktop usage. Datacenters have plenty of raw performance, they need security, so they trade (excessive) performance for security.
      żAbysmal performance? Please, explain me that point because I recognize that the lack of block pointer rewrite is a hit on the performance (that's being worked out) on low space situations ... but tools like external ZIL or cache devices makes a worlds of difference in performance when you have multiple devices (SSDs included) in your hands.

      btrfs doesn't have anything like this (even if we ignore that we are talking about an unstable filesystem) ... if I'm mistaken, please provide me proper information.

      I use ZFS on my desktop, and I can assure you that my computer is not a datacenter ... some people use it even in lower end hardware and more constrained situation with no issues.
      Of course, you won't exploit the full potential of ZFS without proper gear; but that's another story that also applies to other filesystems

      Yes and no ... It's a port, correct.
      But platorm is properly abstracted and the "core" of the filesystem is VERY portable that's why the feature flags were introduced in the first place.

      http://svnweb.freebsd.org/base?view=...evision=236884

      Agreed.
      But we are talking about whole worlds of differences between btrfs and ZFS when it comes to testing.

      Yes ... it was conceived 11 years ago aprox.

      While the ZFS community may care about solving this problem, it's not the highest priority for Sun's customers and, therefore, for the ZFS team.
      That's something from the Sun era ... the development of ZFS changed a lot since that time.
      Have something more recent?

      Licensing discussions, don't really like them ... short story is YES, you're right.


      Regards

      Comment


      • #18
        Originally posted by pankkake View Post
        It is very true that ZFS isn't very good for desktop usage.
        Why, because of memory usage?

        Comment


        • #19
          Originally posted by ryao View Post
          Solaris has the best implementation and FreeBSD has the best port, but ZFS is by no means Solaris-exclusive. There are dozens of operating systems that support it if you consider each distribution to be a different OS. In specific, the following major operating system families have some level of ZFS support:

          Darwin
          FreeBSD
          Linux
          NetBSD
          *** Edit: this one is wrong here **** Solaris/Illumos
          Windows NT
          These are PORTS.
          ZFS was developed for Solaris and so its Solaris exclusive FS, same as XFS to Iris. There are many PORTS of XFS, including port to Linux, but it communicates via interface, which is Iris specific. It is not rewritten FS for, say, Linux. Same ZFS.
          Original claim to post I answered was "Oh, Linux sucks because it doesn't have ZFS" [.. and BSD rocks because it has ZFS].
          The essence is that ZFS was developed for and is truly native only on Solaris. I have nothing against ports btw.

          Originally posted by ryao View Post
          Would you name some?
          Click the link in post please.

          Originally posted by ryao View Post
          Do these limitations include Denial of Service? I have managed to hang systems that use a combination of ext4, CFQ and discard. I have yet to hang a system using ZFS outside of situations that involve experimental patches.
          Never claimed that ZFS is DoS secure, nor stated that BTFS is riper that ZFS.

          Originally posted by ryao View Post
          The same could be said for Linux, yet people who use Linux on desktops would disagree. Similarly, people who use ZFS on desktops (such as myself) would disagree. Having used ZFS on my desktop, I consider it to be the best filesystem available for systems that run mainstream Linux distributions and I strongly encourage its use.
          How much hundred terrabytes your cluster has? Or you use it due to bit-rot protection? Whats the reason you prefer slower FS, compared to EXT4 and BTRFS with LZO?

          Originally posted by ryao View Post
          Would you elaborate on that? I have written dozens of patches to improve the ZFS Linux port that resulted in changes to the upstream repository. All of them were developed on my desktop.
          So you are ZFS port developer and this is your point to have ZFS on Linux? What is your view on ZFS license? Why don't you develop BTFS instead?

          Originally posted by ryao View Post
          What is your point? The GPL is purposely incompatible with a wide range of licenses. In the case of the CDDL, the incompatibility only affects distribution of a kernel binary containing ZFS.
          CDDL was designed to be incompatible with GPL. GPL is much older. There is no point in reinventing same license, but with just a small difference - exactly to be able to sabotage GPL ans claim GPL is incompatible.
          ZFS is ideal flagship for this. Once Sun went down, its stupid engineers decided to take revenge a la DirectX vs OpenGL and arm the ZFS with bomb license.
          If Sun really cared, they would dual-license or grant exception, but no - they purposely take revenge .. against own stupidity.

          Originally posted by ryao View Post
          Would you elaborate? People usually tell me that using ZFS as their rootfs makes their computers perform faster. That has been my experience.
          I have not tested ZFS vs BTRFS personally, because I am more than comfortable with EXT4.
          The only way to prove "faster" claim is to do PTS tests.

          Originally posted by ryao View Post
          How can btrfs be more universal than ZFS when it is limited to Linux?
          BTRFS is free to port away, just as ZFS, and it does not have stupid CDDL license. This makes it more universal.

          Comment


          • #20
            Originally posted by crazycheese View Post
            BTRFS is free to port away, just as ZFS, and it does not have stupid CDDL license. This makes it more universal.
            Please correct me if I'm wrong but to my understanding BTRFS is heavily tied to Linux. This is due to fact that it uses a lot of the already existing Linux specific functionality instead of implementing those itself. ZFS then again does almost everything itself and therefore is more easily portable. I doubt we will ever see BTRFS on any other operating system a side from Linux.

            Comment


            • #21
              Originally posted by vertexSymphony View Post
              Then make up your mind with this statement:

              Note: I added the bold/underline

              ZFS was conceived in Solaris (d'oh !), but not exclusive of this platform ... even if you "only ported to BSD" is still wrong, you say the opposite you said before with the "exclusive".

              I SEE WHAT YOU DID THERE
              I didn't do anything - read my response to first post please.
              The whole "battle" with you rised due to "is/was" and I spelled it wrong - ZFS was developed on Solaris (was exclusive).
              Everywhere I said "ZFS is Solaris exclusive" I referred to its legacy, NOT that one can use ZFS only on Solaris(!).
              Again check my first post as response to first comment.

              Originally posted by vertexSymphony View Post
              Not counting the different *BSD systems.
              I prefer to talk about the community and free ZFS, not the Oracle's one ... ty very much
              So community ZFS differs from "official"? Forked? Which license? How about Oracle patents on ZFS?

              Originally posted by vertexSymphony View Post
              First of all, citing a benchmark from Michael Larabel is something I always "take with tweezers"
              Michael is well known for his "well" and badly done benchmarks mostly because he had no formation on what he was benchmarking ... I remember a really horribly made benchmark (in fact, if my memory doesn't fail in one of the "benchmarks" he didn't use the same hardware) that ended up in a flame war in the mailing lists and this wiki page was born from that: http://wiki.freebsd.org/BenchmarkAdvice

              That article FROM 2010 lacks information about OS, configuration, specific filesystem configurations and some extra information to actually extract something meaninful from these numbers ... sorry, can't take that seriously, but those are cute colored graphs nonetheless
              When it comes to linux graphical stack benchs, some people here also remembers horribly done benchmarks ...
              One should definately rebench XFS, EXT4, BTRFS and ZFS (under proper configuration).
              I have nothing against.
              I am not ZFS user, so I can't help, except on my own. Can do this if everyone denies, because I too have a bit interest.

              Originally posted by vertexSymphony View Post
              żAbysmal performance? Please, explain me that point because I recognize that the lack of block pointer rewrite is a hit on the performance (that's being worked out) on low space situations ... but tools like external ZIL or cache devices makes a worlds of difference in performance when you have multiple devices (SSDs included) in your hands.

              btrfs doesn't have anything like this (even if we ignore that we are talking about an unstable filesystem) ... if I'm mistaken, please provide me proper information.

              I use ZFS on my desktop, and I can assure you that my computer is not a datacenter ... some people use it even in lower end hardware and more constrained situation with no issues.
              Of course, you won't exploit the full potential of ZFS without proper gear; but that's another story that also applies to other filesystems
              WE need fair benchmarking on common desktop hardware.

              Originally posted by vertexSymphony View Post
              Yes and no ... It's a port, correct.
              But platorm is properly abstracted and the "core" of the filesystem is VERY portable that's why the feature flags were introduced in the first place.
              Nice!
              But remember, "port" or not, my sentence had different meaning! We drift too much in meaningless direction.
              I don't care if its port or not, seriously. I am against stupidity as in 1st post in this thread - regardless of OS.

              Originally posted by vertexSymphony View Post
              Agreed.
              But we are talking about whole worlds of differences between btrfs and ZFS when it comes to testing.
              Yes.

              Originally posted by vertexSymphony View Post
              That's something from the Sun era ... the development of ZFS changed a lot since that time.
              Have something more recent?
              Originally posted by vertexSymphony View Post
              Licensing discussions, don't really like them ... short story is YES, you're right.
              Thats the only point that disturbs me on ZFS.
              I know that both are opensource and I don't care about licenses, because I am not developing them.
              My problem is that ZFS license caused integration type like installing proprietary Nvidia driver. :/

              Originally posted by vertexSymphony View Post
              Regards
              Regards..

              Comment


              • #22
                Originally posted by crazycheese View Post
                How much hundred terrabytes your cluster has? Or you use it due to bit-rot protection? Whats the reason you prefer slower FS, compared to EXT4 and BTRFS with LZO?
                Both of those filesystems are slower than ZFS. The main reasons for that are the ARC page replacement algorithm and write sequentialization.

                Originally posted by crazycheese View Post
                So you are ZFS port developer and this is your point to have ZFS on Linux? What is your view on ZFS license? Why don't you develop BTFS instead?
                The CDDL has facilitated inter-operating system development in ways that are unprecedented for the GPL. It is one of the reasons that ZFS development is doing so well.

                With regard to btrfs, you could say the same about <insert filesystem here>. With that said, btrfs is perhaps the most over-hyped filesystem software in the Linux community. btrfs suffers from unbound internal fragmentation, which I consider to be a fatal design flaw. ZFS does what we need, lacks the flaws inherent to btrfs' design and is available on numerous platforms. The only reasons that I can see to develop btrfs are hype, NIH syndrome and attempts to lock people into Linux. None of those things interest me.

                Originally posted by crazycheese View Post
                The only way to prove "faster" claim is to do PTS tests.
                The Phoronix Test Suite tests suffer from stratification bias by measuring workloads that represent a small fraction of real-world usage. Even when they appear to be measuring something that might be useful, I cannot rule out the possibility that I missed a mistake that Michael made given all of the mistakes that I catch in his other benchmarks. I would not rely on them.

                Originally posted by crazycheese View Post
                BTRFS is free to port away, just as ZFS, and it does not have stupid CDDL license. This makes it more universal.
                The CDDL license is more friendly to non-GPL operating systems than the GPL license is.
                Last edited by ryao; 12-15-2012, 05:35 PM.

                Comment


                • #23
                  Originally posted by ryao View Post
                  Both of those filesystems are slower than ZFS. The main reasons for that are the ARC page replacement algorithm and write sequentialization.
                  Originally posted by ryao View Post
                  The Phoronix Test Suite suffer from stratification bias by measuring workloads that represent a small fraction of real-world usage. Even when they appear to be measuring something that might be useful, I cannot rule out the possibility that I missed a mistake that Michael made given all of the mistakes that I catch in his other benchmarks.
                  Got numbers?

                  Originally posted by ryao View Post
                  The CDDL has facilitated inter-operating system development in ways that are unprecedented for the GPL. It is one of the reasons that ZFS development is doing so well.
                  Originally posted by ryao View Post
                  The CDDL license is more friendly to non-GPL operating systems than the GPL license is.
                  Yes, GPL protects open software and is not happy when its consumed by/into proprietary. Full stop.
                  Here comes CDDL, claims its the good license, sabotages GPL and guys like you malform it into GPL causes problems form.
                  Don't do crap. Will be no problems. Pretty easy.

                  The development of proprietary systems/applications or their achievements does not interest me.

                  Originally posted by ryao View Post
                  With regard to btrfs, you could say the same about <insert filesystem here>. With that said, btrfs is perhaps the most over-hyped filesystem software in the Linux community. btrfs suffers from unbound internal fragmentation, which I consider to be a fatal design flaw. ZFS does what we need, lacks the flaws inherent to btrfs' design and it is available on numerous platforms. The only reasons that I can see to develop btrfs are hype, NIH syndrome and attempts to lock people into Linux. None of those things interest me.
                  Its not hype, its working. It has online defrag and internal fragmentation was just a bug, not a flaw. Numbers here say different things. It also does not bring bullshit license with it. Bonuses all around.
                  Also, if this is taken into account, Sun and ZFS can kiss my rear pipe - so much "interest" I have in them, and "available numerous platforms" as well.
                  If you can't use GPL, consider exchanging Gentoo with something more "numerous", "unprecendented" and "closed" - love/hate won't do you good.

                  Comment


                  • #24
                    a comparison matrix between btfrs and zfs

                    Has Phronix produced the titled comparison matrix.

                    It would be nice to know about performance and more than performance, such as system administration management , and any other evaluations.

                    Comment


                    • #25
                      There is no need to sabotage the restrictive GPL even more. It does well by itself.

                      Comment


                      • #26
                        Originally posted by Cthulhux View Post
                        There is no need to sabotage the restrictive GPL even more. It does well by itself.
                        Nope, only with external help, and its still unaffected. Guess, ability to troll from your server minority OS makes you feel better. That damn Linux and GPL, due to them BSD never made it to desktop and got BSDM'ed through Apple. All by itself. Ow, wait..

                        Comment


                        • #27
                          Originally posted by crazycheese View Post
                          Got numbers?
                          Figuring put how to produce meaningful benchmark numbers is something of an art. ZFS collects statistics on cache hits and misses, which would be a good place to start. I often see a >97% cache hit rate with ZFS, which is probably the most important metric in desktop workloads. At the moment, I lack a readily accessible method of getting similar statistics for other filesystems. If you are interested in that, I suggest that you implement it.


                          Originally posted by crazycheese View Post
                          Yes, GPL protects open software and is not happy when its consumed by/into proprietary. Full stop.
                          Here comes CDDL, claims its the good license, sabotages GPL and guys like you malform it into GPL causes problems form.
                          Don't do crap. Will be no problems. Pretty easy.

                          The development of proprietary systems/applications or their achievements does not interest me.


                          Its not hype, its working. It has online defrag and internal fragmentation was just a bug, not a flaw. Numbers here say different things. It also does not bring bullshit license with it. Bonuses all around.
                          Also, if this is taken into account, Sun and ZFS can kiss my rear pipe - so much "interest" I have in them, and "available numerous platforms" as well.
                          If you can't use GPL, consider exchanging Gentoo with something more "numerous", "unprecendented" and "closed" - love/hate won't do you good.
                          The GPL can be just as proprietary and closed in the case of TiVo, Facebook and others (e.g. secure boot). Being open is not one of its merits and quite frankly, I neither believe claims that it can accomplish that nor do I think that its attempts to toward that are a good thing. If someone is going to have a closed platform, then it would be best if the software is well written. If I or a loved one were in a hospital for instance, I would want the medical monitors to be use reliable software. The same goes for flight traffic control systems and other vital infrastructure. I definitely would not want to prioritize some kind of pointless crusade so that I could tell people about how good I am over having reliable infrastructure. I doubt you would either. You could always say that if people did things X way, that would not be a problem. The problem with that is that we would not have any concept of software licensing if people did things X way. Hence, why such discussions are pointless.

                          Comment


                          • #28
                            Originally posted by crazycheese View Post
                            Nope, only with external help, and its still unaffected.
                            It is not.

                            Originally posted by crazycheese View Post
                            Guess, ability to troll from your server minority OS makes you feel better.
                            So, minority OS is a con for you? You use the OS with the largest market share then?

                            Comment

                            Working...
                            X