Re: Problem in accessing executable files

Re: Problem in accessing executable files

[Vineet Joglekar]

Hi John,



Thanks for suggesting the single / few bytes encryption test. I tried doing that, but in vain. Maybe I am going wrong somewhere else.



I will briefly tell the functions I have written and the sequence if I am doing any mistake in the logic, please let me know.

In file.c I have added the info about the functions my_generic_file_read and my_generic_file_write in ext2_file_operations.



For decrypting, the sequence is:

my_generic_file_read ---> my_do_generic_file_read ---> my_file_read_actor ---> my_decrypt_data



I have not made any changes in my_generic_file_read  and my_do_generic_file_read. In the function my_file_read_actor, I copy the page to my buffer (1 page - allocated at the time of mounting) I decrypt that buffer and pass it to the __copy_to_user() function.



for encrypting, the sequence is:

my_generic_file_write ---> my_encrypt_data



In function my_generic_file_write, between functions __copy_from_user() and commit_write(), I am calling my_encrypt_data() by passing address of the page that is passed to __copy_from_user()



My encrypt / decrypt routine is very basic at this time - just xoring every byte of the page as:

*to = *to ^ 0xff; *to++;



If I change my encrypt/decrypt routines to encrypt / decrypt just first or last byte of the page, then I get a different error saying the file is not executable - when I try to execute it. I thought there might be a problem with executable header, but I guess when I encrypt last byte of the page, header should have been bypassed.



Is it something like, for executables, the data is refered in some other functions - that is, before do_generic_file_read geting called?



Thanks and regards,



Vineet



 --- On Tue 02/01, John T. Williams < jtwilliams@vt.edu > wrote:

From: John T. Williams [mailto: jtwilliams@vt.edu]

To: vintya@excite.com, linux-kernel@vger.kernel.org

     Cc: linux-c-programming@vger.kernel.org

Date: Tue, 1 Feb 2005 10:37:30 -0500

Subject: Re: Problem in accessing executable files



This is just a thought.<br><br>Text files are better able to handle small faults. ie an extra space or<br>characters or even an unreadable piece of data might not cause the file to<br>become unreadable by most text editors.  Binary files aren't as flexible.<br>Every bit could be an instruction to the processor and might cause a seg<br>fault.<br><br>Just to test the theory, I would start by making the encrypt decrypt<br>function only effect the first byte.  If doing this doesn't cause a seg<br>fault, I would recheck my decrypt encrypt algorithm to make sure it doesn't<br>pad or expand at any point. maybe use them on a regular file and the an<br>md5sum on the file before and after, just to make extra sure.<br><br><br>----- Original Message ----- <br>From: "Vineet Joglekar" <vintya@excite.com><br>To: <linux-kernel@vger.kernel.org><br>Cc: <linux-c-programming@vger.kernel.org><br>Sent: Tuesda
 y, February 01, 2005 8:58 AM<br>Subject: Problem in accessing executable files<br><br><br>><br>> Hi all,<br>><br>> I am trying to add some cryptographic functionality to ext2 file system<br>for my masters project. I am working with kernel 2.4.21<br>><br>> since the routines do_generic_file_read and do_generic_file_write are used<br>in reading and writing, I am decrypting and encrypting the data in the resp.<br>functions. This is working fine for regular data files. If I try to copy /<br>execute executable files, I am getting segmentation fault. In kernel<br>messages, I see same functions (read and write) getting called for the<br>executables also. If I comment encrypt/decrypt functions, its working fine.<br>><br>> Now since it is working for regular text files, I suppose there is not a<br>problem in my encrypt/decrypt routines, then what might be going wrong?<br>><br>> Thanks and regar
 ds,<br>><br>> Vineet<br>><br>> _______________________________________________<br>> Join Excite! - http://www.excite.com<br>> The most personalized portal on the Web!<br>> -<br>> To unsubscribe from this list: send the line "unsubscribe<br>linux-c-programming" in<br>> the body of a message to majordomo@vger.kernel.org<br>> More majordomo info at  http://vger.kernel.org/majordomo-info.html<br><br><br>

_______________________________________________
Join Excite! - http://www.excite.com
The most personalized portal on the Web!

Re: Problem in accessing executable files

[Glynn Clements]

Vineet Joglekar wrote:

> Thanks for suggesting the single / few bytes encryption test. I tried
> doing that, but in vain. Maybe I am going wrong somewhere else.

I'm pretty sure that mmap'ed access doesn't go through
do_generic_file_read().

Executable files (programs and shared libraries) are normally mmap'ed
rather than read, which would explain why you have problems with
executable files.

Try writing a test program which reads the contents of a file using
mmap() rather than read(). If my assumption is correct, the decryption
code will be bypassed, so you will get the encrypted data.

-- 
Glynn Clements <glynn@gclements.plus.com>

Re: Problem in accessing executable files

[Vineet Joglekar]

Hi all,



1 more interesting observation regarding my executable file problem.



If I copy an executable say "prac" from ext2 fs to my encrypted fs as prac1, prac1 doesnt run on the encrypted fs. but, if I make another copy, from prac1 to normal ext2 fs, as prac2, then the prac2 executes properly on normal file system. that means my encryption, decryption functions are getting executed properly while using that executable as normal file to read and write, but something else is happening when I try to execute it. I am failing to understand what am I missing.



Vineet



 --- On Sun 02/06, Vineet Joglekar < vintya@excite.com > wrote:

From: Vineet Joglekar [mailto: vintya@excite.com]

To: jtwilliams@vt.edu, linux-kernel@vger.kernel.org

     Cc: linux-c-programming@vger.kernel.org

Date: Sun,  6 Feb 2005 17:41:01 -0500 (EST)

Subject: Re: Problem in accessing executable files



<br>Hi John,<br><br>Thanks for suggesting the single / few bytes encryption test. I tried doing that, but in vain. Maybe I am going wrong somewhere else.<br><br>I will briefly tell the functions I have written and the sequence if I am doing any mistake in the logic, please let me know.<br>In file.c I have added the info about the functions my_generic_file_read and my_generic_file_write in ext2_file_operations.<br><br>For decrypting, the sequence is:<br>my_generic_file_read ---> my_do_generic_file_read ---> my_file_read_actor ---> my_decrypt_data<br><br>I have not made any changes in my_generic_file_read  and my_do_generic_file_read. In the function my_file_read_actor, I copy the page to my buffer (1 page - allocated at the time of mounting) I decrypt that buffer and pass it to the __copy_to_user() function.<br><br>for encrypting, the sequence is:<br>my_generic_file_write ---> my_encrypt_
 data<br><br>In function my_generic_file_write, between functions __copy_from_user() and commit_write(), I am calling my_encrypt_data() by passing address of the page that is passed to __copy_from_user()<br><br>My encrypt / decrypt routine is very basic at this time - just xoring every byte of the page as:<br>*to = *to ^ 0xff; *to++;<br><br>If I change my encrypt/decrypt routines to encrypt / decrypt just first or last byte of the page, then I get a different error saying the file is not executable - when I try to execute it. I thought there might be a problem with executable header, but I guess when I encrypt last byte of the page, header should have been bypassed.<br><br>Is it something like, for executables, the data is refered in some other functions - that is, before do_generic_file_read geting called?<br><br>Thanks and regards,<br><br>Vineet<br><br> --- On Tue 02/01, John T. Willi
 ams < jtwilliams@vt.edu > wrote:<br>From: John T. Williams [mailto: jtwilliams@vt.edu]<br>To: vintya@excite.com, linux-kernel@vger.kernel.org<br>     Cc: linux-c-programming@vger.kernel.org<br>Date: Tue, 1 Feb 2005 10:37:30 -0500<br>Subject: Re: Problem in accessing executable files<br><br>This is just a thought.<br><br>Text files are better able to handle small faults. ie an extra space or<br>characters or even an unreadable piece of data might not cause the file to<br>become unreadable by most text editors.  Binary files aren't as flexible.<br>Every bit could be an instruction to the processor and might cause a seg<br>fault.<br><br>Just to test the theory, I would start by making the encrypt decrypt<br>function only effect the first byte.  If doing this doesn't cause a seg<br>fault, I would recheck my decrypt encrypt algorithm to make sure it doesn't<br>pad or expand at any point. ma
 ybe use them on a regular file and the an<br>md5sum on the file before and after, just to make extra sure.<br><br><br>----- Original Message ----- <br>From: "Vineet Joglekar" <vintya@excite.com><br>To: <linux-kernel@vger.kernel.org><br>Cc: <linux-c-programming@vger.kernel.org><br>Sent: Tuesday, February 01, 2005 8:58 AM<br>Subject: Problem in accessing executable files<br><br><br>><br>> Hi all,<br>><br>> I am trying to add some cryptographic functionality to ext2 file system<br>for my masters project. I am working with kernel 2.4.21<br>><br>> since the routines do_generic_file_read and do_generic_file_write are used<br>in reading and writing, I am decrypting and encrypting the data in the resp.<br>functions. This is working fine for regular data files. If I try to copy /<br>execute executable files, I am getting segmentation fault. In kernel<br>messages, I see same functions (read and 
 write) getting called for the<br>executables also. If I comment encrypt/decrypt functions, its working fine.<br>><br>> Now since it is working for regular text files, I suppose there is not a<br>problem in my encrypt/decrypt routines, then what might be going wrong?<br>><br>> Thanks and regards,<br>><br>> Vineet<br>><br>> _______________________________________________<br>> Join Excite! - http://www.excite.com<br>> The most personalized portal on the Web!<br>> -<br>> To unsubscribe from this list:
  send the line "unsubscribe<br>linux-c-programming" in<br>> the body of a message to majordomo@vger.kernel.org<br>> More majordomo info at  http://vger.kernel.org/majordomo-info.html<br><br><br><br><br>_______________________________________________<br>Join Excite! - http://www.excite.com<br>The most personalized portal on the Web!<br>-<br>To unsubscribe from this list: send the line "unsubscribe linux-c-programming" in<br>the body of a message to majordomo@vger.kernel.org<br>More majordomo info at  http://vger.kernel.org/majordomo-info.html<br>

_______________________________________________
Join Excite! - http://www.excite.com
The most personalized portal on the Web!

Re: Problem in accessing executable files

[Al Viro]

On Tue, Feb 08, 2005 at 07:14:27PM -0500, Vineet Joglekar wrote:
> 
> Hi all,
> 
> 
> 
> 1 more interesting observation regarding my executable file problem.
> 
> 
> 
> If I copy an executable say "prac" from ext2 fs to my encrypted fs as prac1, prac1 doesnt run on the encrypted fs. but, if I make another copy, from prac1 to normal ext2 fs, as prac2, then the prac2 executes properly on normal file system. that means my encryption, decryption functions are getting executed properly while using that executable as normal file to read and write, but something else is happening when I try to execute it. I am failing to understand what am I missing.

1) 6 or 7 newlines
2) mmap()

Re: Problem in accessing executable files

[Vineet Joglekar]

Hi Glynn,



Thanks a lot for the reply. I will try that too.



I forgot to mention 1 thing, that I had put printk() in the do_generic_file_read function to see if it is getting executed, and I saw the printed message for executables too. Thats why I assumed that these functions do get executed while accessing executable files. Is it someting like other mmap functions get executed before calling regular read?



Thanks and regards,



Vineet 



 --- On Mon 02/07, Glynn Clements < glynn@gclements.plus.com > wrote:

From: Glynn Clements [mailto: glynn@gclements.plus.com]

To: vintya@excite.com

     Cc: jtwilliams@vt.edu, linux-c-programming@vger.kernel.org

Date: Mon, 7 Feb 2005 18:31:24 +0000

Subject: Re: Problem in accessing executable files



<br>Vineet Joglekar wrote:<br><br>> Thanks for suggesting the single / few bytes encryption test. I tried<br>> doing that, but in vain. Maybe I am going wrong somewhere else.<br><br>I'm pretty sure that mmap'ed access doesn't go through<br>do_generic_file_read().<br><br>Executable files (programs and shared libraries) are normally mmap'ed<br>rather than read, which would explain why you have problems with<br>executable files.<br><br>Try writing a test program which reads the contents of a file using<br>mmap() rather than read(). If my assumption is correct, the decryption<br>code will be bypassed, so you will get the encrypted data.<br><br>-- <br>Glynn Clements <glynn@gclements.plus.com><br>

_______________________________________________
Join Excite! - http://www.excite.com
The most personalized portal on the Web!

Problem in accessing executable files

[Vineet Joglekar]

Hi all,



I am trying to add some cryptographic functionality to ext2 file system for my masters project. I am working with kernel 2.4.21



since the routines do_generic_file_read and do_generic_file_write are used in reading and writing, I am decrypting and encrypting the data in the resp. functions. This is working fine for regular data files. If I try to copy / execute executable files, I am getting segmentation fault. In kernel messages, I see same functions (read and write) getting called for the executables also. If I comment encrypt/decrypt functions, its working fine.



Now since it is working for regular text files, I suppose there is not a problem in my encrypt/decrypt routines, then what might be going wrong?



Thanks and regards,



Vineet

_______________________________________________
Join Excite! - http://www.excite.com
The most personalized portal on the Web!

Re: Problem in accessing executable files

[John T. Williams]

This is just a thought.

Text files are better able to handle small faults. ie an extra space or
characters or even an unreadable piece of data might not cause the file to
become unreadable by most text editors.  Binary files aren't as flexible.
Every bit could be an instruction to the processor and might cause a seg
fault.

Just to test the theory, I would start by making the encrypt decrypt
function only effect the first byte.  If doing this doesn't cause a seg
fault, I would recheck my decrypt encrypt algorithm to make sure it doesn't
pad or expand at any point. maybe use them on a regular file and the an
md5sum on the file before and after, just to make extra sure.


----- Original Message ----- 
From: "Vineet Joglekar" <vintya@excite.com>
To: <linux-kernel@vger.kernel.org>
Cc: <linux-c-programming@vger.kernel.org>
Sent: Tuesday, February 01, 2005 8:58 AM
Subject: Problem in accessing executable files


>
> Hi all,
>
> I am trying to add some cryptographic functionality to ext2 file system
for my masters project. I am working with kernel 2.4.21
>
> since the routines do_generic_file_read and do_generic_file_write are used
in reading and writing, I am decrypting and encrypting the data in the resp.
functions. This is working fine for regular data files. If I try to copy /
execute executable files, I am getting segmentation fault. In kernel
messages, I see same functions (read and write) getting called for the
executables also. If I comment encrypt/decrypt functions, its working fine.
>
> Now since it is working for regular text files, I suppose there is not a
problem in my encrypt/decrypt routines, then what might be going wrong?
>
> Thanks and regards,
>
> Vineet
>
> _______________________________________________
> Join Excite! - http://www.excite.com
> The most personalized portal on the Web!
> -
> To unsubscribe from this list: send the line "unsubscribe
linux-c-programming" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html