Armory - Discussion Thread

[Ploo]

Redownloaded windows-all and 64 bit for good measure. Uninstalled and removed Armory as you said and reinstalled - same issue. The 64-bit obviously gave a 'not valid win32 app' error.

This is frustrating - really want to get an offline wallet going. Maybe I should man up and just use ubuntu.

While I do generally approve of Ubuntu/Linux, in general, that doesn't change the fact that the App should work on Windows XP.  I'm not sure what could be causing it...

Maybe try installing MSVC++ Redistributable 2008.  Perhaps that will put a usable msvcp90.dll file into your system32 dir.  (you can copy it into the Armory directory to be sure).  I really going out on a limb, though...

I also googled the error and came to a conclusion I was missing MSVC++ but I downloaded the 2010 redistributable. I got the 2008 and it works, thanks!

[etotheipi]

I also googled the error and came to a conclusion I was missing MSVC++ but I downloaded the 2010 redistributable. I got the 2008 and it works, thanks!

Wow, I can't believe that worked!  I only suggested 2008 because I compile with MS Visual Studio 2008, so it seemed like a natural fit.  But...

(1) I don't know why 2010 wouldn't work
(2) I don't know why that fixed it, since most users don't have the redistributable installed yet Armory still works on their system...!?!

Bizarre.  But I'm glad it works!  I'll add that to my list of things to recommend for Windows users with problems running Armory!

I just got an error.
I have one adress with two tx inside (= I sent 1 and 2 bitcoins there). Trying to send the whole sum of one of the input transactions (2 bitcoins), Armory wants to send fees. I accept sending fees, and tell Armory to send the remaining back to the same adress. Result:

SelectCoins returned a list of size zero. This is problematic and probably not your fault.

There are a dozen ways to avoid this, as a user. But since you asked for testers and feedback, I will post everything here, until you ask me to use bitcoin-qt again ;-)

edit: tx went through when the fee and bitcoins to be sent were smaller than one input.

Gah!  I thought coin-selection was really stable by now.  If this happens again, and you're willing to send me your watching-only wallet, please let know.  I really want to catch that error in the debugger.  I've had a couple reports of this in the past year, and each time I've implemented something that fixed it for that user.  I guess there's still strange conditions that can cause it...

Btw, yes:  this is exactly the stuff I want out of testers.  There are so many conditions, and so many things that can break from just trivial-looking one-line code changes.  It freaks me out even reorganize code or delete comments before a release, for fear of accidentally changing code flow and breaking seemingly-unrelated things...

[etotheipi]

Oh, a timelock! I like that idea! :-)
"Dear heir. It seems like I was hit by a bus. Sorry. Please search long enough to find the right five-letter password for a little surprise."

I actually had thought about that for the Android phone encryption -- the phone will be slow, and I don't want to drain the battery by using some crazy long key stretching.  But even a "weak" key stretching will give you enough time to move the coins before someone could reasonably brute force it (24H, if they're good)

Maybe that's a good compromise.  It would also go along with the idea of having an encrypted test phrase with the wallet, so you outsource the brute-forcing without give them access to the wallet directly (you give them test-encrypted data, they can prove when they've found the encrypted passphrase by giving you the decrypted data/string, but they can't take your coins until you work out an arrangement with them to exchange passphrase for BTC)

[chrisrico]

I'm really excited!

After considering the criticism of my Armory Pi setup, I spent the day learning serial ports, (more) python, and Armory. I'll release the codes tomorrow, but my solution is pretty sweet. I disabled the serial terminal and replaced it with a program that listens for unsigned transactions. When it sees one, it unlocks the wallet with the passphrase entered on the Pi's own keyboard, signs the transaction, and sends it back over the serial port. The other modification is to Armory, specifically the first offline transaction screen. Along with "Save as file" and "Copy to clipboard" options, there is now a "Transmit over serial" button, which does just what it seems. When it receives the signed transaction back, it automatically advances to and fills the text box on the transaction broadcast screen.

All this required surprisingly little code, but I need some sleep right now!

edit... I found a bug (or two). If you're running the LevelDB version of Bitcoin, Armory won't go online if the blk00*.dat files don't exist, even if the blocks/blk000*.dat files do. Also, I saw a lot of this

Code:

***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...

at what I am assuming is the end of a block data file, before the next one begins scanning. It can last for quite a few seconds sometimes, churning those out 10 per seconds or so.

[picobit]

I do not own a safe, so if someone were to break in my house and get to the paper backup, the funds would probably be gone before I would realized it had been stolen. So my idea is to have hdd backup of the wallet with a strong passphrase, and a single copy of a paper wallet with a weak enough passphrase that I could bruteforce it in a decent amount of time in case I would forget it, but would give me enough time to clear the funds out if someone were to steal it. I hope that makes sense.

That would  be a most unusual burglar.  99% of burglars would say "crap, a piece of paper, not money, not drugs ... worthless".  The last percent would actually look on the paper before saying "what a nerd" and throw it away.

[Ente]

I'm really excited!

After considering the criticism of my Armory Pi setup, I spent the day learning serial ports, (more) python, and Armory. I'll release the codes tomorrow, but my solution is pretty sweet. I disabled the serial terminal and replaced it with a program that listens for unsigned transactions. When it sees one, it unlocks the wallet with the passphrase entered on the Pi's own keyboard, signs the transaction, and sends it back over the serial port. The other modification is to Armory, specifically the first offline transaction screen. Along with "Save as file" and "Copy to clipboard" options, there is now a "Transmit over serial" button, which does just what it seems. When it receives the signed transaction back, it automatically advances to and fills the text box on the transaction broadcast screen.

All this required surprisingly little code, but I need some sleep right now!

Awesome! That's the way! :-)
I was thinking about how to automatize the whole external signing, but didn't find such a nice solution like you!
That way it would be almost-fully-automatized, with only one click extra and one passphrase on pi's keyboard!
I like it!
So, is the Pi available again? :-)

Ente

[prezbo]

I do not own a safe, so if someone were to break in my house and get to the paper backup, the funds would probably be gone before I would realized it had been stolen. So my idea is to have hdd backup of the wallet with a strong passphrase, and a single copy of a paper wallet with a weak enough passphrase that I could bruteforce it in a decent amount of time in case I would forget it, but would give me enough time to clear the funds out if someone were to steal it. I hope that makes sense.

That would  be a most unusual burglar.  99% of burglars would say "crap, a piece of paper, not money, not drugs ... worthless".  The last percent would actually look on the paper before saying "what a nerd" and throw it away.

I would be more inclined to think like this if the paper backup didn't have "Armory Wallet" written in bold on it If I were a burglar and saw a piece of paper with the word wallet on it, I would surely look into it.

[chrisrico]

Here is a diff of my changes: http://pastebin.com/A9x0grtz

There's probably a better way to do some of the threaded stuff for the listener, and I'll probably clone the Github project and submit a push request to make it easier for either etotheipi or others to fix it up, but right now it works pretty well!

[etotheipi]

I'm really excited!

After considering the criticism of my Armory Pi setup, I spent the day learning serial ports, (more) python, and Armory. I'll release the codes tomorrow, but my solution is pretty sweet. I disabled the serial terminal and replaced it with a program that listens for unsigned transactions. When it sees one, it unlocks the wallet with the passphrase entered on the Pi's own keyboard, signs the transaction, and sends it back over the serial port. The other modification is to Armory, specifically the first offline transaction screen. Along with "Save as file" and "Copy to clipboard" options, there is now a "Transmit over serial" button, which does just what it seems. When it receives the signed transaction back, it automatically advances to and fills the text box on the transaction broadcast screen.

All this required surprisingly little code, but I need some sleep right now!

Yes!  I agree this is an excellent solution -- one that I very much wanted to implement myself, but decided could not be promoted as a general solution -- due to the risk of users not disabling terminal logins via serial (thus, actually decreasing security for a number of impatient/inattentive users). 

But, I always wanted such a solution that could be documented on the website as "Use at your own risk, but it's a pretty sweet solution if you do it right...".  I would personally use this, I just never got around to implementing a serial protocol.  I suppose, alternatively, I could find a way to make an Armory offline bundle DVD -- with OS and everything preconfigured correctly for this purpose.  Then the serial solution would be perfect  

Until then, I will graciously accept the modifications to Armory as a side-branch, or maybe a diff file or something that can be downloaded/distributed separately.  Put instructions on the website with appropriate warnings.

[chrisrico]

Yes!  I agree this is an excellent solution -- one that I very much wanted to implement myself, but decided could not be promoted as a general solution -- due to the risk of users not disabling terminal logins via serial (thus, actually decreasing security for a number of impatient/inattentive users). 

But, I always wanted such a solution that could be documented on the website as "Use at your own risk, but it's a pretty sweet solution if you do it right...".  I would personally use this, I just never got around to implementing a serial protocol.  I suppose, alternatively, I could find a way to make an Armory offline bundle DVD -- with OS and everything preconfigured correctly for this purpose.  Then the serial solution would be perfect  

Until then, I will graciously accept the modifications to Armory as a side-branch, or maybe a diff file or something that can be downloaded/distributed separately.  Put instructions on the website with appropriate warnings.

Thanks! Yeah, I don't think this needs to be a core part of Armory, especially since there are so few changes necessary, and Armory is very easy to build (at least on Linux).

I've run into one strange issue though... the extra scripts work just fine on my online machine, but on the Pi they stopped working. It's not giving me an "unrecognized symbol: armoryengine error", it just doesn't recognize CLI_ARGS (or any other "constant" from armoryengine), even though I'm doing "from armoryengine import *". Oddly, running "python armoryengine.py" works just fine, and starting up a REPL and doing "from armoryengine import *" lets me "print CLI_ARGS" without issue. Any ideas?

edit... Hmm. It must be a path issue of some kind, because when I move the script out of the extras directory and remove the "sys.path.append('..')", it works.

[marcus_of_augustus]

After considering the criticism of my Armory Pi setup, I spent the day learning serial ports, (more) python, and Armory. I'll release the codes tomorrow, but my solution is pretty sweet. I disabled the serial terminal and replaced it with a program that listens for unsigned transactions. When it sees one, it unlocks the wallet with the passphrase entered on the Pi's own keyboard, signs the transaction, and sends it back over the serial port. The other modification is to Armory, specifically the first offline transaction screen. Along with "Save as file" and "Copy to clipboard" options, there is now a "Transmit over serial" button, which does just what it seems. When it receives the signed transaction back, it automatically advances to and fills the text box on the transaction broadcast screen.

Okay, this raises the bar significantly. Basically, you have built yourself a RasbPi minimally-connected 'hardware' wallet for Armory ... good stuff.

[etotheipi]

edit... I found a bug (or two). If you're running the LevelDB version of Bitcoin, Armory won't go online if the blk00*.dat files don't exist, even if the blocks/blk000*.dat files do. Also, I saw a lot of this

Code:

***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...
***ERROR:  parseNewBlockData did not get enough data...

at what I am assuming is the end of a block data file, before the next one begins scanning. It can last for quite a few seconds sometimes, churning those out 10 per seconds or so.

SOB!  I even moved my blk files around to test this before committing the upgrade to beta.  If this truly is broken, I need to fix that, because I was under the assumption that this should coast users through the 0.8 upgrade.  

And given those errors you report, looks like it may be more than renaming block files.  Perhaps they changed the way data is written, too...?  Gah!  I guess I need to just download 0.8 and test it myself...

[chrisrico]

SOB!  I even moved my blk files around to test this before committing the upgrade to beta.  If this truly is broken, I need to fix that, because I was under the assumption that this should coast users through the 0.8 upgrade. 

And given those errors you report, looks like it may be more than renaming block files.  Perhaps they changed the way data is written, too...?  Gah!  I guess I need to just download 0.8 and test it myself...

Sorry, I might not have been clear. It reads the files, but when it gets to the "end" of one, it throws this error a bunch (random amount?), before it starts reading the next.

[etotheipi]

SOB!  I even moved my blk files around to test this before committing the upgrade to beta.  If this truly is broken, I need to fix that, because I was under the assumption that this should coast users through the 0.8 upgrade.  

And given those errors you report, looks like it may be more than renaming block files.  Perhaps they changed the way data is written, too...?  Gah!  I guess I need to just download 0.8 and test it myself...

Sorry, I might not have been clear. It reads the files, but when it gets to the "end" of one, it throws this error a bunch (random amount?), before it starts reading the next. However, it does seem like maybe once it catches up, Armory doesn't update when new blocks come in when connected to a 0.8 node.

Is there a compiled version of 0.8, or do I have to build it myself?  All my updates for 0.8 were based on the assumption that nothing is changing except for the locations and names of the block files.   At least, that's all the core developers told me about

[chrisrico]

In this thread Peter posted a link to some experimental pre-0.8 builds on his server. That's what I'm running. Block data files are now ~128MB, numbered blk000X.dat, and in the .bitcoin/blocks folder. Other than that, I'm not sure what changed.

Oh, and it looks like there is a problem where Armory doesn't see any new blocks coming in after it's caught up, when connected to a 0.8 client. I thought it was a permissions issue, but after fixing that I'm still seeing it. Nothing interesting in the log, either.

[chrisrico]

Piggybacking off my other posts, here are the changes I made to the Pi:

In /etc/inittab, change the line that looks like this

Code:

1:2345:respawn/bin/agetty ...

to this

Code:

1:2345:respawn:/bin/login -f pi tty1 </dev/tty1 >/dev/tty1 2>&1

This will automatically log in as the pi user to the first tty device upon boot.

Also, remove or comment out the line that looks like this

Code:

T0:23:respawn:/sbin/getty -L ttyAMA0 115200 vt100

That disables logging in over the serial port.

In /boot/cmdline.txt, remove console=ttyAMA0,115200 kgdboc=ttyAMA0,115200 from

Code:

dwc_otg.lpm_enable=0 console=ttyAMA0,115200 kgdboc=ttyAMA0,115200 console=tty1

This disables outputting boot and debug information over the serial port.

Modify /etc/sudoers (with visudo) to include this line

Code:

pi ALL=NOPASSWD: /sbin/poweroff

This lets you you power off the pi without entering your sudo password.

Finally, in /home/pi/.bashrc, add the following at the bottom

Code:

python $armory_directory/offline_serial_server.py /dev/ttyAMA0 9600
sudo poweroff

Voila. Your Pi will now boot into the offline serial server, allowing you to communicate with it from Armory using the patch I posted at the top of the page. When you're finished, hit CTRL-C and the Pi will cleanly power down.

[etotheipi]

In this thread Peter posted a link to some experimental pre-0.8 builds on his server. That's what I'm running. Block data files are now ~128MB, numbered blk000X.dat, and in the .bitcoin/blocks folder. Other than that, I'm not sure what changed.

Yeah, I know what they said it was changing to, that's where this function came from: findLatestBlkFiles().  It should default to looking for blocks/blk0000X.dat, and if those aren't there, switch to blk000X.dat.  It's really annoying that the old version is 4 digits and 1-indexed, the new version is 5 digits and zero-indexed (accommodating both versions is ugly, oh well.)

Thanks for the link.  I'll dig in when I get some time after RAM reduction (which is going mediocre at the moment, but I may have something worth testing in the next week or two).  

[BRules]

tried to find the answer but couldn't. Is there a way to verify a message signed in armory with the bitcoinqt and vice versa?

[Ploo]

Here's something that's on my mind that I hope to get cleared up.

I currently have a cold-storage set up thanks to Armory. Something that only just now came to my attention is the fact that the online watch-only wallet is managing to generate addresses without knowing their private keys. How does it do this? Does the watch-only copy derive a new seed that is only capable of generating public keys in the series? What's the process behind this? Or does the watch-only copy come with a healthy bulk of pre-generated addresses?

[chrisrico]

Does the watch-only copy derive a new seed that is only capable of generating public keys in the series? What's the process behind this?

That's exactly it. Here is the code that creates a watching-only fork of the wallet.

Basically, the root key has a public and private portion. Given only the public portion, you can derive the subsequent public keys, but not the private keys, for the same reason that you can't derive a private key from a public one.