I am attempting to learn more about C and it's descendants(C++ mainly). I have decided that I would like to create a "file system" of sorts. Not a particularly advanced one mind you but something to play with. I have no intentions of making it mountable, securable or even recoverable.

At the moment I am stuck in concept land with trying to decide how to implement the MFT/FAT.

At first I thought that I would just use the first X number of bytes to store a lookup table, when I realized that there would then be a limitation to the number of files I could store I thought maybe use some type of metadata with each file but then I would have to scan the entire filesystem to locate a file.

I have read through this and this although the z80 link seems like it is more up my alley.

From a high level I want to be able to issue a command like:

./myfs funnycat.jpg mystorage.mfs

Essentially appending binary data to the end of mystorage.mfs

How can I store the information that would contain the file names, start offset and length while avoiding the issue of self imposed limits (lookup table length) or having to scan the entire file (metadata with binary data)?

Concise Explanation I am looking for a way to label binary data stored in a single contiguous file so that I can pull data from a given offset range or by string.

./myfs mystorage.mfs funnycat.jpg

Likely in order to accomplish this I will add some logic to myfs to check the first argument for signs that it is a blob containing other files or not.

  • How to "best" store information that an FS requires is a spectacular question that would revolutionize all of computing if anybody had the answer to it. As it is however, we have numerous file systems in existence with varied approaches to these issues that all have benefits and caveats throughout them. People could give you many ideas and suggestions here, but the best way to do what you describe? People have been designing file systems for decades trying to come up with the best way, and none have seen certain success yet. Commented May 21, 2014 at 17:52
  • I'm close voting as too broad right now, you need to narrow your scope by saying what approach you plan to do so that we can talk about what problems it has and potential solutions to those problems. Right now it just openly asks for any approaches to solve your problem which just feels too broad to be answerable. Commented May 21, 2014 at 17:53
  • @JimmyHoffa Agreed the wording is a bit misleading, but it reads to me like he just wants a simple metadata scheme for finding something in a binary blob that won't impose unreasonable hard-coded limits on the mock file system.
    – Doval
    Commented May 21, 2014 at 18:02
  • @Doval you have hit the nail on the head here. JimmyHoffa I have made a small edit to trim down the scope.
    – xandout
    Commented May 21, 2014 at 18:09
  • @Doval yeah I thought it over, made a tiny edit and retracted my close vote. Commented May 21, 2014 at 18:17

1 Answer 1


Below are some possible solutions to some of the problems have mentioned.

How do I store a number of arbitrary size?

A simple approach is to use a Variable Length Quantity. Basically, for each octet used to represent your number, 7/8 of its bits are used to represent the number and the extra bit is set to 1 if there are any more octets. You can also just pick some "unreachable" size and write code to support this fixed size.

How can I tell when I've reached the end of a file?

You can use some form of file allocation table (you can represent it as a linked list to allow it to grow to arbitrary size).

I want to support disk fragmentation. I.e., I should be able to grow a file, even if the file is about to overlap another file, without either file.

Have each file be made up of clusters of some fixed size. Each cluster will have a header with information like:

  • How much of the cluster is allocated? (clusters are fixed size, a cluster's content may not be fully populated with data).
  • Where is the next cluster? (A file may need more than one cluster. You could store this information in an allocation table instead, of course).
  • Is this cluster allocated? (If this information is in the cluster rather than in an allocation table, you'll be forced to "format" your entire "hard drive" using a "slow format." Using a file allocation table allows you to perform a "quick format." Mind you, you could cheat by leveraging your existing file system (i.e., you can assume byte is initialized to 0).

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.