OSDev.org

What do i need to do to find a file with a specified name on a fat12 partitioned disk in the boot sector

Read the root directory into memory and search for the file with the correct name. If you want to load that file, follow the cluster chain in the FAT.

Microsoft published a specification for FAT12 (and FAT16 and FAT32) as part of the EFI specification. Here's a copy of that specification converted to PDF. You can download the original Word document directly from Microsoft.

I've written code to load a file from a FAT12 partition on a hard disk. You can find it here if you're interested. (I've made a few updates recently that haven't yet made it onto Github... nothing important, though; just changing the load address.)

I don't think that is possible within the limited space of the boot sector. You need to load a second stage loader (from fixed sectors) in the boot sector, and then load the file in the second stage loader.

Sure it is! See http://www.brokenthorn.com/Resources/OSDev6.html

OK, it might be possible but it is still not a good idea. By keeping stage2 in the root directory you risk it getting deleted or corrupted, so a better idea is to place it in reserved sectors directly after the boot sector. Additionally, you might decide to format your boot partition as FAT32, EXT or NTFS, and then you need a new boot sector to load the second stage, and it might not be possible to handle reading the root directory of EXT or NTFS from the boot sector.

Everything is in the root directory! By that logic, you shouldn't store any part of your OS as files on the disk because that risks it getting deleted or corrupted.


That's why EXT2 and NTFS have more than one sector for the boot "sector".

That's why EXT2 and NTFS have more than one sector for the boot "sector".[/quote]
Well, you say that, but my previous attempt to write a real-mode ext2 file loader finally clocked in at 2048 bytes. Unfortunately, ext2 only gives me 1024 bytes of space (irrespective of block size, by the way).

However, those 1024 bytes are enough to load i-node 5 (which is the ext2 reserved i-node for the boot loader). So if I had gone that route, I would have had routines to load sectors in the first 512 bytes, routines to load i-nodes in the second 512 bytes, then skip 1024 bytes for superblock and block group descriptors, then routines to traverse directories in however much memory was needed.

But I never finished it. My main goal was to only use the first 1024 bytes. But it seems like that is just not enough space for

• enabling A20-gate
• testing if the CPU is acceptable
• loading second bootloader block and superblock
• traversing directories to find a programmable path name
• finally loading that one to the 1MB line
• mapping it as an ELF file
• doing the rest of stuff my kernel needs
• and finally transitioning to 64-bit mode

I don't think anyone is suggesting that you do all of those things within the EXT2 boot block; just finding and loading the second stage is enough.

Not in my design. I can have several bootable images in the root directory, and if one is deleted or corrupt I can always boot the other. That's why I call the second one for "safe". However, the second stage loader only exists in a single instance, and it is never modified, so it's better placed in reserved sectors.

For UEFI boot, things are a bit more safe since the loaders & images are on it's own UEFI system partition.

You can have a separate boot partition on legacy BIOS too.

Late to the party, but here is also a fat32 implementation which loads a file in the "stage1": https://bitbucket.org/AlexOSAdventurer/alo_chainloader/src/master/src/boot_stage1/ . This is if you're wanting something that will work with a large fat system. I heavily commented it to death too so it should be straightforward to understand the process.