Counterfeit Xiaomi Phone – An Intriguing Deception

I figured I’d use this post to detail a matter that’s still under investigation by eBay regarding a seller located in either Singapore or China (supposedly China, but the package had a Singapore EMS number) kindly providing me with a clearly fake Xiaomi Mi3 cellular phone. What’s interesting about this is not so much the fact that counterfeit or otherwise questionable items come from China – we all know that it’s pretty much a risk inherent in doing any sort of business in Asia, and I certainly wouldn’t have even considered making the purchase were I not assured that eBay or PayPal would have my back should the seller do anything dodgy. No, the real story here is how damn good the fakes have become and why a novice user may not even be able to spot that anything is at all amiss. Certainly from a security perspective there is potential here – who the hell knows what’s buried in their firmware. The organization have clearly demonstrated their lack of ethics by producing a counterfeit device, so what other little secrets is this device hiding? Here’s some information – and I do admit that I am being kind of lazy here and using parts of this report as my eBay complaint detail.

The first clue when unboxing the device was that the back of the chassis did not feel exactly like the coated aluminum/magnesium chassis of the original phone – but this is far too subjective to be useful to a first time buyer. I noticed that the USB cable which shipped with the device had a “MI” logo on it – something which the genuine USB cable does not have. The charger is unquestionably of the same style that lacks adequate shielding and was responsible for electrocuting several users a while back, and has MI branding screenprinted onto it in a manner that the OEM simply wouldn’t tolerate (heck, it wasn’t even printed straight!).

The counterfeit USB cable features "MI" branding while the official one does not.

The counterfeit USB cable features “MI” branding while the official one does not.

When booting the phone I noticed that the “Optimizing” screen featured a stock Android jellybean image and not the anime style bunny screen that Xiaomi use for their Mi products (they actually sell these bunnies on their e-store. I can’t see why anyone would love their device so much they have to have a plush logo, but maybe it is cultural). Later I found that the Fastboot and bootloader screens are also missing the Mi branding, something which other users who have discovered fakes and Xiaomi themselves have revealed is a dead giveaway of a clone.

The fastboot is a generic jellybean Android icon whereas the official Mi3 features an anime style bunny.

The fastboot is a generic jellybean Android icon whereas the official Mi3 features an anime style bunny.

This is what the fastboot screen is supposed to look like!

This is what the fastboot screen is supposed to look like!

The screen is another point of contention. If you take a screenshot and then check the properties of the image generated, you’ll see that it has a resolution of 720×1280. This is a problem seeing that the Mi3 has a resolution of 1080×1920. The screen also lacks the oleophobic Gorilla Glass III that the official phone has.

The resolution of the display is 720x1280 and not the expected 1080x1920.

The resolution of the display is 720×1280 and not the expected 1080×1920.

Very few benchmarking applications will run successfully on the phone and are force-closed, presumably by some kind of watchdog script to ensure that users don’t easily discover the truth of what lies inside their device. In the stock “About” screen everything appears in order. One application that did run successfully, hwinfo provides another clue to the schizophrenic nature of the beast – two mutually exclusive product codes are used in the same screen, “pisces” (the TD-CDMA version for China Telecom) and “cancro” (the Qualcomm based WCDMA international version). It also notes an incorrect physical screen size and states a PPI of 320 dpi (official phone reports and is spec’d at 441). If you place the device on an accurate scale you’ll also find that it is several grams off the official weight, something which can’t easily be explained away (hint: the included battery is much smaller than the quoted 3050mah cell).

Both "pisces" (TD-CDMA) and "cancro" (WCDMA) showing in Hardware Info.

Both “pisces” (TD-CDMA) and “cancro” (WCDMA) showing in Hardware Info.

Entering fastboot mode and issuing a “getvar product” to the device reveals its dirty little secret. This device does not contain the Snapdragon 800 MSM8974, rather the Snapdragon S2 era MSM7627A. Geekbench 3, the only benchmarking app I could actually execute successfully on the device reveals in its multi-core test that the device scored a 806, putting it only slightly above the GNex. Other Snapdragon 800 devices scored much, much higher – for example the LG Nexus 5 at 2538 and the Kindle Fire HDX7 at 2730. The single core comparison was even more pathetic, with the device scoring 295 (worse than the Moto G’s Snapdragon 400). It also turns out the reported memory of 2GB is also a downright fabrication, with actual device memory being reported by Geekbench as 843MB – clearly there is only a 1GB module on board.

Fastboot displays MSM7627A instead of MSM8974*

Fastboot displays MSM7627A instead of MSM8974*

The log file from Xiaomi's flashing tool, MiFlash shows the flash failed due to wrong product ID.

The log file from Xiaomi’s flashing tool, MiFlash shows the flash failed due to wrong product ID.

The processor doesn't even remotely approach the scores of a genuine  Snapdragon 800.

The processor doesn’t even remotely approach the scores of a genuine Snapdragon 800.

The memory reported by Geekbench does not appear to correlate with the 2GB specified by Xiaomi.

The memory reported by Geekbench does not appear to correlate with the 2GB specified by Xiaomi.

Xiaomi has an app that is designed to test for fakes, but unfortunately it is Chinese. Nevertheless I ran the device, which reported the device was not a confirmed Xiaomi. I also checked the serial number on the website, which was genuine but had been checked many times, indicating that they are simply cloning “official” serial numbers and issuing them to multiple devices. The IMEI reported to the BTS and the IMEI displayed by the software (e.g. by *#06#) are also different, with the displayed one agreeing with the serial on the box. Obviously they knew that they had to ensure that the IMEI looked unique at least to the cellular network provider lest their little scam be immediately detected by a provider.

Xiaomi's own testing application highlighting in red the device's benchmark versus the expected score in black.

Xiaomi’s own testing application highlighting in red the device’s benchmark versus the expected score in black.

Xiaomi's testing tool highlighting hardware discrepancies.

Xiaomi’s testing tool highlighting hardware discrepancies.

As you’d expect the official Xiaomi firmware will not flash, either using the included updating app or via fastboot. The flashing tool checks the processor model prior to flashing, which obviously returns the incorrect MSM7627A and the flash does not proceed. This is yet another dead giveaway. Another clue is the performance of the phone, which is sluggish at best and lacks sufficient grunt to even decode some audio files in real time without stuttering when multitasking. The barometric pressure sensor provides a constant output of 1013.* mbar with the last digit moving randomly as if the sensor is actually present. Unfortunately this doesn’t even remotely correlate with the area QNH. There is definetely a GPS module present, but fixes take several minutes and are only obtainable and maintained when literally holding the device horizontally in the air.

Perhaps the most disappointing thing about this fake is that the WCDMA module only supports 2100Mhz (official phone supports quad band WCDMA). This makes it useless in our locale except in certain areas where coverage is lacking and my provider uses 2100Mhz mini-cells to enhance coverage in built up areas such as city centers. Everywhere else it unfortunately falls back to GSM, and often cannot even maintain a GPRS or EDGE connection for more than a few brief moments. If you are speaking during handoff to another cell, the phone invariably reboots. It actually seems like it reboots more often than it stays operational. I ensured I was in a 2100mhz serviced area to do these tests, as it was very unstable both without a SIM and on GSM.

Another interesting point is what I found on the embedded flash of the phone – some screenshots in Russian appearing to highlight the same kind of inconsistencies I am speaking of. Evidently the seller attempted to sell the phone to a Russian buyer, who returned the phone after discovering it was a counterfeit and I have been handed his phone (yet another eBay policy violation from the seller, who stated the device was Brand New). An interesting thing the Russian guy noticed and clearly wanted to make evident in one of his captures was the resolution of a captured photo was just 6 megapixel at 16:9 (the phone specs state 13MP).

One of seven Russian language screenshots highlighting inconsistencies.

One of seven Russian language screenshots highlighting inconsistencies.

So, there you have it. Perhaps the worst thing will be waiting for my refund, and perhaps having to pay postage, etc to return this item to the seller. He has already offered me $70 to simply walk away and accept this piece of junk. I don’t think that is going to be happening any time soon.

NB: The only edits made to photography were to redact the cellular network name from the top display of Android in the spirit of this blog never posting personally identifiable information either of my own self or of contributing authors.

Lenovo Superfish: Not The First Time Vendors Have Preinstalled Malware

Aside

The recent fiasco involving Lenovo and what has come to be known as Superfish that amongst other things performs a MiTM attack on TLS encrypted HTTP traffic so that it can still insert advertisments into encrypted pages.

The story broke on Feb 18 when major tech sites and even mainstream media began speaking of a security issue affecting recent laptops from Chinese vendor Lenovo. Given the large amount of negative publicity the company has reacted quite swiftly an has even provided a list of affected laptops and portables along with an easy to use Superfish removal tool. LastPass has a third party tool which will quickly identify the presence of Superfish on your machine. Given that it does not actively hide its presence, there are likely far more straightforward ways to identify the presence of the malware.

Of course this isn’t the first time OEMs have saddled their customers with preinstalled junk that is either annoying or worse compromises the confidentiality of any data on that machine. An excellent example would be the multiple vendors who include a variety of Popcap games which include the perennially problematic OpenCandy advertising engine and trial copies of software that you have no realistic chance of ever requiring let alone purchasing on a whim.

I have a big problem with Microsoft Windows, particularly 8x so any time I purchase a new machine I typically do the following:

  • Boot the machine, skipping the sysprep/OOBE screens so that i may enter Device Manager and note down the hardware within the machine (generally done only if the unit’s manual does not list full specifications and there is no reliable information online).
  • Reboot whilst pressing the boot selection hotkey so that I can boot the clonezilla DVD, which is relatively lightweight and includes the tools we need. Obviously you could substitute any similarly equipped Linux, *BSD or Solaris live CD/DVD.
  • Using hdparm(8) a master password is set on the HDD and if the disk appears frozen, the power connector is temporarily disconnected. An ATA SECURE ERASE (enhanced) is then executed. This ensures that there is no chance of any data remanence (particularly important if your new PC is ex-display and there have been myriad of people playing with it in the store).
  • Following the conclusion of the ATA SECURE ERASE command, smartctl(8) is used to run a long self-test and the outcome of this is noted.
  • I go ahead with the installation of FreeBSD or Debian Linux.

Canary Page Updated (Probably Leaving) and I’m Still Alive

Hi everyone! I’ve been away for quite a while. Work and family life have got the best of me, and I simply haven’t had much time to do anything remotely enjoyable – and well, authoring articles for this blog falls into the latter bucket. I have leave coming soon and intend to spend at least some of that time catching up with the people over at Bruce Schneier’s website and those who have commented here and/or e-mailed me. I apologize for the delay.

The canary has been updated and re-signed less someone think I’ve been taken captive by the CIA and forced to eat rodents for dinner. I am considering doing away with the canary page due to its limited utility, I make a brief but considered statement about this on the canary page itself.

Millions of Google Credentials Leaked

Almost five million gmail credentials were posted to a Russian language bitcoin forum a few days ago. Google’s official position is that these credentials were harvested externally and that gmail itself does not have a vulnerability that has allowed credential disclosure.

This seems like a reasonable position given the limited (relative to the massive number of gmail users) number of credentials leaked. I believe the list was harvested through the use of password stealing malware or through social engineering (e.g. phishing) and/or a combination of such techniques. Some of the leaked passwords appeared unlikely to have been dictionary cracked so a leak of hashed passwords from Google looks even more unlikely.

UPDATE: A site has appeared – isleaked.com (unfortunately also Russian language) to allow concerned users to search the list of leaked credentials to see if they are affected. I would not personally enter my gmail address into a web resource published by an unknown party, as while the author’s intent may be benevolent it is also equally likely that form submits are being harvested for unsolicited email lists.

UPDATE: PC World and the mainstream tech media have taken up the story.

Matt Green on PGP’s Shortcomings

Matt Green recently blogged about the shortcomings of PGP for e-mail encryption. He makes some valid points, and without a doubt the trust management of PGP and its clone GNUPG is probably its Achillies’ heel. The “web of trust” was supposed to counter the issues inherent in heirarchical certification authority schemes like X.509, and for the most part it does a reasonable job at doing just that when the number of group participants are small. In the real world it suffers from much of the same human factors that have brought the CA style model into question over the past decade. There isn’t an easy answer to this ongoing engineering problem and until a reliable, decentralized way of establishing identity is developed. I suspect that the ultimate solution will draw inspiration from current cryptocurrency “proof of work” type systems.

Updated Canary

Seeing as I have unfortunately been away from this blog for a while I figured it best that I update my canary document so that nobody need concern themselves that I have been compromised. I will endeavor to return to providing a high quality commentary on the current matters of concern within our industry within the following few weeks as my personal situation slowly returns to normal.

My Apologies For The Brief Hiatus

Aside

I have been away from this blog and most of my other responsibilities for a little over a week as a result of tending to some family issues and preparing ourselves for a move across town.

Fortunately I will be back on track in the next few days and will update this blog again very shortly.