Last October, I learned about “key stretching” from Solar Designer’s “How to manage a PHP application’s users and passwords” — a concept that is applicable well beyond PHP. Combined with salting, it is effective at slowing down brute forcing attacks, and even against rainbow tables.
When CTRL-C in gdb shuts down a program insted of interrupting it
According to The Linux Programming Interface, a well-behaved multi-threaded UNIX program should use sigwait() or sigwaitinfo() instead of signal() or sigaction(). A linux-only program could even use signalfd().
Unfortunately, Linux/UNIX programs using sigwait() are hard to interrupt in the debugger. Instead of interrupting the program, it terminates the program. How does one work around this problem? See my answer on stackoverflow.
C++ initializer lists
One of the cool new features in C++ is initializer lists. Here’s an example from Wikipedia:
struct Object
{
float first;
int second;
};
Object scalar = {0.43f, 10}; // One Object, with first=0.43f and second=10
Object anArray[] = {{13.4f, 3}, {43.28f, 29}, {5.934f, 17}}; // An array of three Objects
There’s another example over at stackoverflow.
IBM has published the useful C++0x feature support in GCC 4.5.
Autojump: Faster than ‘cd’
Here’s a cool tool: autojump, written by Joel Schaerer (thanks, Joel). I spend much of my day as a programmer navigating around in the linux filesystem. Built-in tools like ‘pushd’ and ‘popd’ are nice, as are subprocesses — e.g.
(cd ~/Download && wget http://somewhere.com/path/to/file)
… and when it finishes downloading, I’m still in the directory I was in before the download was started.
Now there’s autojump to add in to the mix. After I ‘cd’ to various directories, later, I can type ‘j Down’ to cd to my Downloads directory. Very convenient. I just wish it were built into every distribution of linux.
Linux tty auditing
Since RHEL 5.4, and in recent Fedora releases, it’s possible to audit what users type at their tty (command prompt), thanks to the work of Steve Grubb, a RedHat employee.
Edit /etc/pam.d/system-auth and append the following, but not both:
session required pam\_tty\_audit.so disable=\* enable=root session required pam\_tty\_audit.so enable=\*
Wait for users to log in and type into a terminal. Later, to see audited tty input, run:
aureport --tty
When a user logs in, the pam module tells the kernel to enable tty auditing for a process and its children. All tty input is logged, but it may not be incredibly easy to read (it includes backspaces, control characters, etc.). I’m unclear as to when and how often the kernel flushes out accumulated tty input to the audit log. The records are identified with a type of TTY in /var/log/audit/audit.log.
In addition to tty auditing, RedHat patched their bash shell so that it neatly audits each and every command line it executes, with a record type of USER\_TTY. It’s prettier to read than raw tty auditing — and it’s easy for a user to bypass by using a shell that doesn’t send its commands to the Linux audit system, like zsh, or a custom-built unpatched bash. Maybe that’s why “aureport –tty” doesn’t show USER\_TTY records.
—-
The Linux auditing system is powerful. It’s possible to write rules that watch for modification to certain files, or that log the use of certain system calls. See the “audit.rules” manpage for more information.
Pitfalls of verifying signed jar files
In the Java world, it’s possible to digitally sign a jar file using ‘keytool’ to generate or import a digital signature, and ‘jarsigner’ to do the signing. What isn’t so obvious is that when we use ‘jarsigner’ to verify a signed jar, it doesn’t verify that we trust the signature that signed the file. It simply tells us whether the contents were signed by a public key that was included with the jar file.
Surprisingly, there’s no option to tell jarsigner to check for trusted signatures.
In code, we can use java.util.jar.JarFile to check the validity of a jar file. By default, the constructor to JarFile says we want to check the validity. Code must then iterate through each entry in the JarFile and seek to the end of each input stream, otherwise, the integrity isn’t checked. In other words, the java.util.jar.JarFile doesn’t give us the integrity checking with a simple method call such as isValid(), and it doesn’t give us an easy way to check that we trust the signature that the entries were signed with.
Anyone, anywhere, can create their own certificate, and sign a jar file — so if we want to establish trust for a signed jar, we get to do extra work. On stackoverflow.com, Jarek Przygódzki linked to code that shows how to check for trusted signatures.
I wonder why establishing trust for a signed jar isn’t easier. Could it be that signed jar files originated in the bygone era when we ran Java applets in our web browsers? Did web browsers use their certificate authority database to verify some level of trust for the signature contained in a jar file?
Verifying trust is a delicate issue, as demonstrated by the recently hacked certificate authorities including Diginotar and Comodo. Perhaps it’s a good thing that Java’s libraries and command line tools don’t make it deceptively simple to check jar files based on certificates trusted by Certificate Authorities.
Still, I wish the documentation for jarsigner and JarFile would shed more light on the limits of their default verification. I’d call it “hash checking” or “integrity checking based on hashing”.
Easy updates with iOS and Linux vs bomardment from Windows
When I want to check email on my iPod Touch, I simply unlock it, load the email application, and read messages. What a joy. The tool works and works for me. It is my servant, instead of me being its servant.
Compare that to my corporate Windows Laptop. I open the lid and unlock it, wait for corporate anti-virus (I think) to chew through the I/O on my machine for several minutes. Then I can read my email using the unimpressive and slow Lotus Notes (I’d rather be using Thunderbird, Outlook, or speedy mutt).
Or compare to my home Windows Laptop. I open the lid, log in, and am bombarded with requests from several programs to install updates. I just want to check my email, not be attacked with annoying “update me!” windows. I see one from Thunderbird, another from Adobe Flash, another from Mozy and another from Firefox. By the time I’ve upgraded, sometimes I’ve forgotten what I wanted to accomplish in the first place.
Apple has it right with updates for the iPod Touch: They stay out of my way until I want to bother with them, and then they’re all manged together.
Linux has it right: A single package manager updates everything, in one shot, including from third parties.
I’ll continue to use Windows. There are other benefits, even though third-party updates are annoying.
The dog ate my grub.conf; unwanted LVM indirection
In September, I updated packages on my Fedora 14 machine using ‘yum’ (Yellowdog Updater Modified), and rebooted. I was greeted with an unfriendly grub prompt. Boot was broken, and I didn’t know why.
Normally when I want to rescue a Linux system, I use the Fedora install DVD’s rescue mode. Unfortunately, my DVD drive stopped working just after the warranty expired. So I booted using a Linux-on-a-USB image that I carry around with me.
I planned on replicating the rescue mode experience by manually mounting the partitions in /mnt/sysimage, then chroot, and then run ‘grub-install’ to fix things. And that’s what I did, except for a few snags.
The first snag is that I had installed Fedora with default partitioning options (something I rarely do), and it had helpfully used LVM to configure my disk drive. LVM adds a layer of indirection, so I had to figure out how to mount an LVM volume instead of a disk partition. As a general rule, I never use LVM on desktop or laptop machines because I don’t like the additional complexity, especially when I’m trying to mount a drive to recover data.
The next snag is that Linux uses UUIDs for hard disks these days, instead of helpful, human-readable labels. In the good old days, Linux distributions would label the root partition, the home partition, and others. But not when I let the installer work its default magic. So, I had to figure out which volume contained my root partition, my boot partition, and so forth.
Once I mounted my partitions properly, I was ready to “chroot /mnt/sysimage”, and after I did, I ran “grub-install”. No luck. This was the third snag.
I realized that /boot/grub/grub.conf was empty. And grub-install doesn’t fix the problem — it can’t populate information that is completely missing, like knowing what disk UUID to use for the root partition.
“No problem,” I thought, “I’ll just restore from backup…”. Except, I didn’t have a backup. As a general rule, I back up /var, /home and /etc, but not /boot. Now I know better.
Surely, I thought, there should be a copy of grub.conf that the package updates had left around. But there wasn’t.
Such a critical file, and there was no automated backup. No “restore point”. No way to recreate it from data in /etc. What an oversight!
This is the first time I can remember having had an update eat my grub.conf file, and I’ve been running Linux since before distributions switched to grub as a boot loader. On the other hand, maybe I jumped to to an incorrect conclusion — it’s possible that the updates didn’t eat my grub.conf. Regardless, grub.conf was empty and my system wouldn’t boot.
It seems to me that something this important should have been designed with more resilence.
There were two ways to fix it, as far as I could see. One is upgrade from Fedora 14 to Fedora 14 (thus preserving my data). The second is to recreate grub.conf by hand, and for that, I needed an example — a template. I used the one at http://forums.fedoraforum.org/showthread.php?t=227358. My boot up screen isn’t as pretty as it once was, but it is functional.
Lastly, I made copy of /boot/grub/grub.conf. Now where did I put it?…
ORM Solutions for C++ and sqlite: hiberlite, SOCI
I think SQLite is a great database for a C/C++ application to use. It sure is easy to query the data that way, without intrinsic support from the application. When the application changes, the database is still accessible via standard sqlite clients. Yet using SQLite in an application can require a lot of repetitive code to serialize objects to and from the database, so I wondered if there are ORM solutions for C/C++. It turns out that there are. I only had to turn to stackoverflow.com to find the following (among many):
NPR: When Patents Attack!
I write software for a living, and I have a hard time recognizing redeeming value of software patents (other than to have something to negotiate with when someone tries to sue a company). Reading the NPR transcript about patents got my blood boiling. The system has been gamed, and patent troll companies legally extort money from other companies.
Read or listen here. Are there good reasons to have software patents, and if so, on what types of inventions and in what circumstances?