[Pharo-dev] Recommendation for password encryption?

On 12 Jul 2013, at 16:44, Norbert Hartl <[hidden email]> wrote:

> Maybe you can specify your problem a bit more so.

I think what he basically means: the app needs to access the db, for which it needs a db password. How should he store it ?

It also depends if there is only one db password, or one for each user, and in the latter case, if it is the same as the one entered by the user.

Am 12.07.2013 um 15:47 schrieb Mariano Martinez Peck <[hidden email]>:

Hi guys. Some time ago I asked about the best way to encrypt a password. Some of the recommendations I got was:

1) Generally for passwords you do not write a decrypt method for security purposes  (eg you do not want people to be able to reverse engineer the encrypted password easily)

2) put a little salt

3) There is http://www.squeaksource.com/Cryptography/PasswordHashingFFI

4) I should not decrypt password but instead I should compare both encrypted versions to see if they are equal.

My scenario is the following. I need to store database passwords in text files. So I don't want to let them as plain text. And while I understand 4), I think it's a different scenario than mine. I know that some database driver may allow you to directly send the encrypted password. But still...I have another similar scenario in which I have to store a password and then send it (unencrypted)...so I kind of really need a bi-directional encryption.

Now..I would try to avoid having an external library + FFI. Any ideas what is the best thing I can do? And how could I solve 1) if I want bidirectional like in my case?  

"I know that some database driver may allow you to directly send the encrypted password".

That would be the same as having no encryption at all. Everyone knowing the encrypted password could authenticate.

I'm not sure I understand the complete requirements. Do you really don't want to store them as plain text files or do you want to prevent others from being able to read them?

Basically a password that is decryptable would be useless because in order to make it secure you would need to store some other information to decrypt it somewhere thus shifting the same problem from the password to the "something else".

Assuming you have encrypted password files I can also assume that the process reading it is to be considered trusted. Otherwise it won't work because the password needs to be in its decrypted form somewhere in this process and this only works if you can trust the process. 

This problem is often solved using filesystem permissions. Often programs are started as root and so they can read files where root is the only one having read permisson. After doing priviledged stuff the program drops its priviledge and acts as a normal user. In case of a pharo image that wouldn't be to easy. You could have a small program started as root reads the information from the file and starts pharo as normal user passing the information in. Or you could start pharo and another priviledged program that reads the password file and sets the information in the pharo image via socket or something similar. Or you are able to start the pharo process with a dedicated user and arrange the filesystem permission in a way that only this user has read access. Basically they are all the same.

If those are no options than it will be hard. Of course you can encypt the password file itself but that won't work without extra measurement. That is always the trap in thinking about security. The encryption of the password file would be possible only if you have something you can decrypt it with. And that information would have exactly the same security implications as the password itself. Thus it solves nothing but only shifts the responsibility for security to another piece of information.  

I want to prevent others from being able to read them. 

 

Basically a password that is decryptable would be useless because in order to make it secure you would need to store some other information to decrypt it somewhere thus shifting the same problem from the password to the "something else".

Yes, but it would make things A LITTLE bit more complicated. One thing is to browse a simple .txt editor with ssh and another one is find the Pharo .image, open the image, search the code that decrypts, decrypt the password etc... So at least it adds some effort to the hacker :)

 

Ok. What I meant is that if you encrypt the password, there is some password2 you need to encrypt the first password. I was assuming you don't want to keep a secret inside the image. And then it would be hard because password2 would need to be stored secure as well. If you are able to keep a secret in the image then this is a good idea and provides a different type of safety than the filesystem stuff. Well, you could add the filesystem protection as well and then you're ok. 

Assuming you have encrypted password files I can also assume that the process reading it is to be considered trusted. Otherwise it won't work because the password needs to be in its decrypted form somewhere in this process and this only works if you can trust the process. 

yes

 

This problem is often solved using filesystem permissions. Often programs are started as root and so they can read files where root is the only one having read permisson. After doing priviledged stuff the program drops its priviledge and acts as a normal user. In case of a pharo image that wouldn't be to easy. You could have a small program started as root reads the information from the file and starts pharo as normal user passing the information in. Or you could start pharo and another priviledged program that reads the password file and sets the information in the pharo image via socket or something similar. Or you are able to start the pharo process with a dedicated user and arrange the filesystem permission in a way that only this user has read access. Basically they are all the same.

mmm i understand. Doesn't look easy. The last one "Or you are able to start the pharo process with a dedicated user and arrange the filesystem permission in a way that only this user has read access" sounds one more thing to do that could help. The only thing this can fail is if the user the hacker use to get inside is the same that runs Pharo or one with root provilegies. 

 

Exactly. If someone is able to gain root priviledge there is not much left you can do.

That is one of the reasons why programs should have no special priviledges or drop them after doing important stuff. Writing a C application that starts your image, read password file and drops priviledge is not as hard as it sounds. Probably the way you pass the information to the image is a bit more.

You certainly can not pass it as an argument to the image without further measurement because it would be visible by programs like ps. But there are ways to hide it or you can you use a pipe.

If those are no options than it will be hard. Of course you can encypt the password file itself but that won't work without extra measurement. That is always the trap in thinking about security. The encryption of the password file would be possible only if you have something you can decrypt it with. And that information would have exactly the same security implications as the password itself. Thus it solves nothing but only shifts the responsibility for security to another piece of information.  

yes, but as I wrote a bit, it will make it a bit more complicated!

 

Thanks Norbert for the ideas. 

My pleasure!

Norbert

 

This problem is often solved using filesystem permissions. Often programs are started as root and so they can read files where root is the only one having read permisson. After doing priviledged stuff the program drops its priviledge and acts as a normal user. In case of a pharo image that wouldn't be to easy. You could have a small program started as root reads the information from the file and starts pharo as normal user passing the information in. Or you could start pharo and another priviledged program that reads the password file and sets the information in the pharo image via socket or something similar. Or you are able to start the pharo process with a dedicated user and arrange the filesystem permission in a way that only this user has read access. Basically they are all the same.

mmm i understand. Doesn't look easy. The last one "Or you are able to start the pharo process with a dedicated user and arrange the filesystem permission in a way that only this user has read access" sounds one more thing to do that could help. The only thing this can fail is if the user the hacker use to get inside is the same that runs Pharo or one with root provilegies. 

 

Exactly. If someone is able to gain root priviledge there is not much left you can do. That is one of the reasons why programs should have no special priviledges or drop them after doing important stuff. Writing a C application that starts your image, read password file and drops priviledge is not as hard as it sounds. Probably the way you pass the information to the image is a bit more. You certainly can not pass it as an argument to the image without further measurement because it would be visible by programs like ps. But there are ways to hide it or you can you use a pipe.

If those are no options than it will be hard. Of course you can encypt the password file itself but that won't work without extra measurement. That is always the trap in thinking about security. The encryption of the password file would be possible only if you have something you can decrypt it with. And that information would have exactly the same security implications as the password itself. Thus it solves nothing but only shifts the responsibility for security to another piece of information.  

yes, but as I wrote a bit, it will make it a bit more complicated!

 

Thanks Norbert for the ideas. 

My pleasure!

Norbert

I want to prevent others from being able to read them. 

 

Basically a password that is decryptable would be useless because in order to make it secure you would need to store some other information to decrypt it somewhere thus shifting the same problem from the password to the "something else".

Yes, but it would make things A LITTLE bit more complicated. One thing is to browse a simple .txt editor with ssh and another one is find the Pharo .image, open the image, search the code that decrypts, decrypt the password etc... So at least it adds some effort to the hacker :)

 

Ok. What I meant is that if you encrypt the password, there is some password2 you need to encrypt the first password. I was assuming you don't want to keep a secret inside the image. And then it would be hard because password2 would need to be stored secure as well. If you are able to keep a secret in the image then this is a good idea and provides a different type of safety than the filesystem stuff. Well, you could add the filesystem protection as well and then you're ok. 

mmmm sorry I didn't get it. password2 would be a "salt" in this case?

you could explain a bit more? 

But imagine that at some point I DO need to decrypt in the image side. Which algorithm should I use in Pharo?

 

Assuming you have encrypted password files I can also assume that the process reading it is to be considered trusted. Otherwise it won't work because the password needs to be in its decrypted form somewhere in this process and this only works if you can trust the process. 

yes

 

This problem is often solved using filesystem permissions. Often programs are started as root and so they can read files where root is the only one having read permisson. After doing priviledged stuff the program drops its priviledge and acts as a normal user. In case of a pharo image that wouldn't be to easy. You could have a small program started as root reads the information from the file and starts pharo as normal user passing the information in. Or you could start pharo and another priviledged program that reads the password file and sets the information in the pharo image via socket or something similar. Or you are able to start the pharo process with a dedicated user and arrange the filesystem permission in a way that only this user has read access. Basically they are all the same.

mmm i understand. Doesn't look easy. The last one "Or you are able to start the pharo process with a dedicated user and arrange the filesystem permission in a way that only this user has read access" sounds one more thing to do that could help. The only thing this can fail is if the user the hacker use to get inside is the same that runs Pharo or one with root provilegies. 

 

Exactly. If someone is able to gain root priviledge there is not much left you can do.

Indeed. So the only particular concern for this case is if it get access to the user than runs Pharo and that has read access to such files. 

 

That is one of the reasons why programs should have no special priviledges or drop them after doing important stuff. Writing a C application that starts your image, read password file and drops priviledge is not as hard as it sounds. Probably the way you pass the information to the image is a bit more.

I understand. 

 

You certainly can not pass it as an argument to the image without further measurement because it would be visible by programs like ps. But there are ways to hide it or you can you use a pipe.

If those are no options than it will be hard. Of course you can encypt the password file itself but that won't work without extra measurement. That is always the trap in thinking about security. The encryption of the password file would be possible only if you have something you can decrypt it with. And that information would have exactly the same security implications as the password itself. Thus it solves nothing but only shifts the responsibility for security to another piece of information.  

yes, but as I wrote a bit, it will make it a bit more complicated!

 

Thanks Norbert for the ideas. 

My pleasure!

Norbert

--
Mariano
http://marianopeck.wordpress.com

Hi Mariano,

You shouldn't use it.  Its broken. I didn't realize until just now.  But I
also haven't been using it.

To encrypt the db password do

enc:=Blowfish encryptString:'myDbPassword' with: 'mySecretKey'

and to decrypt it later do

Blowfish decryptToString: enc with: 'mySecretKey'

But if you decrypt it with the key '1234' you get

'ë "~ýãîfword'

which shows its only encrypting the first 8 bytes.  And if your db password
is 30 chars long like this one:

'123456789012345678901234567890' then the leaked info is the last 22
numbers.

The tests aren't covering this error yet.  I'll mess with it sometime soon
and get it sorted but for now its broken and shouldn't be used.  Unless your
DB password is 8 bytes or less.








--
View this message in context: http://forum.world.st/Pharo-dev-Recommendation-for-password-encryption-tp4698499p4698682.html

Sent from the Pharo Smalltalk Developers mailing list archive at Nabble.com.

Blowfish is a 8 byte block cipher so for shorter strings I'll need to pad
the byte array, and for longer strings I'll need to make it use cipher block
chaining:
(https://en.wikipedia.org/wiki/Block_cipher_modes_of_operation#Cipher-block_chaining_.28CBC.29)

If you change #decryptString:with: to:

Blowfish class>>decryptString: aString with: aKey
|decryptedData |
decryptedData := (self new ecbDecrypt: aString asByteArray with: aKey
asByteArray  ).
        ^String fromByteArray:  decryptedData asByteArray .

Then this workspace code should work:

| enc encryptedString dkey decrString |
key:='mySecretKey'.
enc:=Blowfish encryptString:'12345678' with: key.

encryptedString := enc asByteArray asString.
Transcript show: ' encrypted:  ', encryptedString; cr.

decrString:=Blowfish decryptString: encryptedString with: key.

Transcript show: ' decrypted:  ', decrString; cr.

But with the password 'test' it will always fail because I'm not yet padding
the byte array to a multiple of 8 bytes before encrypting it.



Thanks for your patience


Paul



--
View this message in context: http://forum.world.st/Pharo-dev-Recommendation-for-password-encryption-tp4698499p4698789.html

Sent from the Pharo Smalltalk Developers mailing list archive at Nabble.com.