Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
Android Bug Security

Every Android Device Launched Since 2012 Impacted By RAMpage Vulnerability (bleepingcomputer.com) 83

Almost all Android devices released since 2012 are vulnerable to RAMpage bug, an international team of academics has revealed today. From a report: The vulnerability, tracked as CVE-2018-9442, is a variation of the Rowhammer attack. Rowhammer is a hardware bug in modern memory cards. A few years back researchers discovered that when someone would send repeated write/read requests to the same row of memory cells, the write/read operations would create an electrical field that would alter data stored on nearby memory. In the following years, researchers discovered that Rowhammer-like attacks affected personal computers, virtual machines, and Android devices. Through further researcher, they also found they could execute Rowhammer attacks via JavaScript code, GPU cards, and network packets.
This discussion has been archived. No new comments can be posted.

Every Android Device Launched Since 2012 Impacted By RAMpage Vulnerability

Comments Filter:
  • by SuperKendall ( 25149 ) on Thursday June 28, 2018 @11:13AM (#56859938)

    The article said that Apple iOS devices may be vulnerable to, but since this is a kind of Rowhammer attack, which Apple mitigated in 2015 [scmagazineuk.com], I wonder if it is?

    It would be nice if the article (and researchers) were more clear about other platforms being vulnerable...

    • this is a kind of Rowhammer attack, which Apple mitigated in 2015 [scmagazineuk.com]

      If Apple says they mitigated it years ago, even without knowing about current research, then we can rest assured that they told the truth just as always, right? Right?

      • this is a kind of Rowhammer attack, which Apple mitigated in 2015 [scmagazineuk.com]

        If Apple says they mitigated it years ago, even without knowing about current research, then we can rest assured that they told the truth just as always, right? Right?

        And Apple cultists with mod points will not stand for any criticism. No, none. Nor have any concept of ethical use of moderation, just like Apple itself has no concept of ethics.

  • by ComputerGeek01 ( 1182793 ) on Thursday June 28, 2018 @11:25AM (#56860004)

    Unless your objective is to crash the device, rowhammer is a useless technique and even then there are far easier ways to accomplish this. Until you can tell me EXACTLY what cells you are modifying and in what way, you will NEVER be able to utilize this vulnerability interesting observation for any kind of useful exploit. Even then, you would have to know WHAT you are modifying and even the most basic memory page protection prevents that. #SLOWNEWSDAY

    • That's exactly my feeling on this as well. It seems much more likely that you could use this to crash a device much more than you could ever use it to perform an attack (since you don't know exactly which cells your memory is in or what the neighbors look like). I'm still not even convinced the Spectre/Meltdown bugs are actually useful in the real world despite all of the hype. To me, a lot of these vulnerabilities boil down to "Oh if you let arbitrary code run, it can run other arbitrary code". Un

      • Unless you can show an exploit that involves a browser visiting a malicious URL in a real-world scenario, I think this is a lot of smoke.

        A number of proofs of concept were published that demonstrated how the exploits could be abused using Javascript. For instance, here's a simple example [react-etc.net] that could be used to provide a script with access to the contents of your memory if your browser/system/chip hasn't been updated to prevent such an attack. This may not impact everyday users much, but if you're on a known, shared system (e.g. AWS, Azure, WordPress, Squarespace, etc.) it becomes far easier to abuse, since you could do something like, say, in

        • ...if you're on a known, shared system (e.g. AWS, Azure, WordPress, Squarespace, etc.) it becomes far easier to abuse, since you could do something like, say, initiate a request that would load a private key into memory, use a sidechannel attack of these sorts to log the memory...

          If you have a private key to load, and you're allowed to do that from javascript, you can just do it directly, there is no need to play games with this. Also, is it true that these services give you some sort of enhanced access where you'd be able to do something special by using your own key from a device that had already logged in with a user account? Wouldn't the key need to be the right key anyways? And if you have my key to load into memory, why would you also need to run the code on my mobile device?

          • If you have a private key to load, and you're allowed to do that from javascript, you can just do it directly, there is no need to play games with this. Also, is it true that these services give you some sort of enhanced access where you'd be able to do something special by using your own key from a device that had already logged in with a user account? Wouldn't the key need to be the right key anyways? And if you have my key to load into memory, why would you also need to run the code on my mobile device?

            I think something got lost in translation, because from what I can tell, very little of this relates back to anything I was talking about.

            [...] and you're allowed to do that from javascript

            I wrongly assumed that it would be clear we weren't talking about Javascript once I switched to discussing AWS and other hosted systems. What language you use actually doesn't matter, so long as you're able to run code on someone else's susceptible system with some guarantees regarding timing. Javascript can be used against everyday users via their browser, but with a sha

            • I wrongly assumed that it would be clear we weren't talking about Javascript once I switched to discussing ...

              No need to read more than that.

              "We" aren't talking about whatever is in your head. Those are thoughts, and other people can't see them; and even if they can infer them, they generally don't want you to presume that your own thoughts rule what "we" were talking about.

              That also explains why it appears you didn't understand my words; you were reading using your own thoughts, instead of using my words.

              And you go right back into these nonsense assertions about being able to read computer memory through a web sit

              • And you go right back into these nonsense assertions about being able to read computer memory through a web site, it is a bunch of "blah blah blah" that doesn't even convince me you know what a website is. You say things like, "Without proper mitigations" and then list things that there were mitigations to protect against back in Netscrape Navigator.

                A) Did you miss that I linked to sample code already? Yes, it was literally possible for the JavaScript on a site to read the contents of your memory by utilizing Spectre (which was mentioned by name earlier in this thread). I’m genuinely befuddled at how someone here doesn’t already understand that and, upon seeing the code I linked, immediately recognize that code as an implementation of such an exploit.

                B) No, these mitigation’s didn’t exist until recently in browsers. Why would th

                • Right, you linked some code, but you don't know what it is. You only know that the tin says something scary.

                  Here, I'll quote you something since you don't know how to do your own research:

                  Here is a quick list of tested browsers and their vulnerability status (always assume the latest version):

                  Firefox -- not vulnerable
                  Firefox ESR -- not vulnerable
                  Internet Explorer 11 -- not vulnerable
                  Microsoft Edge -- not vulnerable
                  Pale Moon -- not vulnerable
                  Waterfox -- not vulnerable
                  Chromium (latest) -- not vulnerable
                  Opera Stable -- not vulnerable
                  Google Chrome Canary -- not vulnerable
                  Google Chrome Stable -- vulnerable*
                  Vivaldi Stable -- vulnerable*

                  *not vulnerable if you enable strict site isolation in the web browser.

                  So if they do their banking in Firefox, then Oh, Snap! you're wrong.

                  • Wait a sec. Let’s back up.

                    Each and every time I referred to a Spectre-based attack via Javascript I very intentionally used past tense or qualified what I said to make it clear I was talking about what had been possible or what could potentially be possible with any systems that hadn’t yet been updated. I never once suggested it was still exploitable via Javascript in any current browser, nor was that something I was ever arguing about with you. My use of present tense was limited to other explo

    • by xanthos ( 73578 )
      Its silly season. With BlackHat and Defcon on the horizon you can expect to see lots of similar "security researcher" announcements over the next few weeks.
    • by bluefoxlucid ( 723572 ) on Thursday June 28, 2018 @11:41AM (#56860126) Homepage Journal

      Pretty much, except for the last bit: memory page protection is administrative, not technical. The CPU says, "You can't write to address 0xC0004000" and you don't. If you write to 0xB0004000 and create electrical fields messing with 0xC0004000, that's physical and bypasses any code that says no you can't.

      It's like approaching a hotel key card lock with dynamite.

      • Comment removed based on user account deletion
        • True. My point was that there isn't page protection for this.

          Some high-criticality stuff, such as disk buffers, use hashing and error protection in software. ECC can get a one-bit-flip correction and two-bit-flip detection per byte (or expanded per word) at 6 bits per byte (75%) overhead. Hamming coding needs an extra two bits whenever the data size doubles, at the expense of taking longer to verify (slower), which means a 4096-bit page only needs 24 bits (3 bytes) of ECC to achieve one-bit repair, and

        • DRAM is getting more expensive these days. It's not just 1/8th more expensive, it's a lot more than 1/8th more power hungry, as you've got to power 9 bits of DRAM for every byte instead of 8, you've for the ECC checking for every read and ECC calculations for every write.
          In saying that, ECC is much more common on SoC's for their caches now than it used to be.

    • it's garbage hardware design, yes.

      electrical engineers have failed in their design of memory and CPU.

      • by bobby ( 109046 )

        it's garbage hardware design, yes.

        electrical engineers' managers have failed in their managment of design of memory and CPU.

        FTFY.

        • No, blame for bad design that causes damages or harm is shared by the engineer, that's how profession works. The engineer who works for PHB that mandates design of bridges that kills shares the guilt for deaths, he was whore who sold himself out.

      • I don't think they failed. They just had different objectives at the time than you would have liked (with 20/20 hindsight). The exploits you mention are based on taking advantage of circuits that were designed for increased performance. If the memory card was more robust to electric field attacks it would have to use redundant or shielded or more spaced out cells (less memory per area) or a stronger supply voltage (more power). You could probably find a way to trade off speed for robustness as well. When yo
      • by sjames ( 1099 )

        Did they? Or did they give the people what they wanted in exchange for a small theoretical risk that has yet to cause a documented real-world issue? You can get ECC RAM that greatly reduces the risk if that's your preference. You can even get radiation hardened equipment that is even less likely to flip a bit if your wallet is up to it.

    • by Anonymous Coward on Thursday June 28, 2018 @12:00PM (#56860234)

      Unless your objective is to crash the device, rowhammer is a useless technique and even then there are far easier ways to accomplish this. Until you can tell me EXACTLY what cells you are modifying and in what way, you will NEVER be able to utilize this vulnerability interesting observation for any kind of useful exploit. Even then, you would have to know WHAT you are modifying and even the most basic memory page protection prevents that. #SLOWNEWSDAY

      Exploits [blackhat.com] have been known since 2015. Basically you fill memory with Page Table Entries, then you corrupt them until a bit flips which gives you access to write your on PTE. From that point, you own the machine. I have not heard of any fixes for this.

    • I am thinking of what you are writing, and then I ask myself "what's the value of the attack",
      a few things of value come into play

      A) the financial calculation of taking out the competitor, design the "attack" as a vendor specific. Enough product problems and they will stop making it and the product will become another Edsel

      B) design the attack to work on specific types of networks that have specific software ( variation of the above ), take for example the security camera's, if the attack can be designed fo

    • by Dwedit ( 232252 )

      People have working root exploits that use Rowhammer, so it's not useless.

    • Plus, my android devices are all so slow that if they were running this type of thing they'd be unusable to me before it even succeeded! They're going to need to run it for weeks, crashing it a zillion times on the way, just to figure out where the memory locations are. They'd never get there without getting wiped.

      • Without kernel ASLR, they can figure out which addresses to hammer at compile time. No need for runtime checking, failing, crashing, etc.
        Kernel physical address aren't going to change much if at all on an Android device since there is usually no swap space. Not sure if ZRAM is common on Android.
        There's a function, virt_to_phys, that converts the virtual address of kernel memory into physical addresses. Handy.

        All they need to do is keep allocating memory or searching through memory their process can write to

        • Without kernel ASLR, they can figure out which addresses to hammer at compile time. No need for runtime checking, failing, crashing, etc.
          Kernel physical address aren't going to change much if at all on an Android device since there is usually no swap space. Not sure if ZRAM is common on Android.
          There's a function, virt_to_phys, that converts the virtual address of kernel memory into physical addresses. Handy.

          All they need to do is keep allocating memory or searching through memory their process can write to until they find an address next to one they want to change.
          They can figure the "next to" bit out by detecting which RAM chips are being used.

          And clown #2 is claiming that banking on Android is safer than on a laptop, because magic unicorns are farting sandboxes.

          So many readers, so few can read! They see something like, "Foo Attack Could Be Really Bad" and they hear, "Foo Attack Is Really Bad." But the words actually imply, "Foo Attack Unlikely To Be Really Bad." You should know that as soon as you see the sensationalist words next to the words that claim ignorance.

          We even have clowns that think that at compile time, maybe their compiler knows th

    • I wish you had researched a bit before posting this. A lot of people modded incorrect information as insightful.
      There are working rowhammer privilege escalations.
      The main one I remember involves mmap()ing a whole asston of anonymous pages, essentially to the point where you fill an appreciable amount of RAM with page table entries. At that point you can use some inference to target those PTEs with good probability of success. As soon as you can modify one PTE to give your process RW access to an arbitrary
    • Until you can tell me EXACTLY what cells you are modifying and in what way, you will NEVER be able to utilize this vulnerability interesting observation for any kind of useful exploit

      And everyone should cast aside their caution and put their faith in Slashdot pronouncements of confident self appointed security expert, exactly why?

  • .. memory chips? If so, then in theory anything using them is vulnerable with enough work. Would using ECC memory avoid all this over hyped crap?
    • by amorsen ( 7485 ) <benny+slashdot@amorsen.dk> on Thursday June 28, 2018 @12:38PM (#56860458)

      Would using ECC memory avoid all this over hyped crap?

      Yes ECC memory and ECC cache mitigates Rowhammer. In theory not completely, you could cause an undetected triple-bit error if you ran the attack long enough. However, in that time you are vastly more likely to hit a detectable-but-uncorrectable two-bit error that halts the machine.

      (A quick Google implied that modern systems are still stuck with single-correction double-detection. I am not sure that is correct.)

      • by bws111 ( 1216812 )

        Or, even better, RAIM.

      • Comment removed based on user account deletion
      • modern systems are still stuck with single-correction double-detection. I am not sure that is correct.

        I am not sure either -- but decades ago it took 5 extra bits to guard 16 bits. As I understood it, the ECC was kept there, and on failure it morphed and used 4 bits to point to the incorrect bit and 1 bit to indicate action needed (ie, failure.)

        This was only 1 vendor from decades ago, but that's what I remember the technical manual saying (back when they described things.) So I assume from that that it takes more bits to correct more bits; that they haven't improved the math behind the detection algorit

        • by amorsen ( 7485 )

          You can't really improve the algorithm, you can only switch which patterns you correct, which you detect, and which you fail on. Provably optimal algorithms have been known basically since anyone cared.

          However, modern memory is 64 bit, not 16 bit. That gives you 20 bit of ECC to play with, and it ought to be possible to do much better with that.

    • The article should be making the distinction.

      It's only a DRAM problem because of DRAM's requirement for refreshing. Other RAM types don't have the constant discharge curve so won't be susceptible.

      SRAM is the fastest. Buffers and caches are built of SRAM. MRAM is waiting in the wings to replace all DRAM and possibly some bulk SRAM in the larger caches too.

      I'm guessing improved DRAM refresh logic can fix the issue though.

Dennis Ritchie is twice as bright as Steve Jobs, and only half wrong. -- Jim Gettys

Working...