Oh god, so much wrong here.
Your stack overlaps your entire code area. And your read buffer. Your read buffer starts at 0x7e00, so that's good, but where it ends is dependent on the BPB fields. With your defaults in the fields, the root directory is 14 sectors. So the root directory buffer is from 0x7e00 to 0x9a00. So you have about 26kB of stack before it clobbers your root directory.
Speaking of that, your algorithm for calculating the number of root directory sectors is inefficient and wrong. You take the number of root directory entries, mutiply them with 32 and divide them with the number of bytes per sector. But that last one is rounding down. Unless your number of root directory entries happens to be divisible by 16, your result will be 1 too low. Also, both the mul and the div can be replaced by a shift. In fact, you can combine the shift count beforehand, so there is no possibility of overflow: "Bytes per sector" must be a power of 2, right? So you can use "bsf" to log2 it, in all cases. Then subtract 5 and you have your right shift count. Take the number of root directory entries, add (1 << count), subtract 1, and >> count the result. That implements the division, but rounding up.
Also, you assume the root directory to be perfectly defragmented. It needn't be; that is sort of the purpose of the whole operation. The pseudo code for finding anything in any directory on any kind of FAT file system is:
Code:
cluster = startcluster; /* read from dir entry or BPB, for the root directory */
while (!found && cluster != endcluster) {
readcluster(cluster); /* handle errors yourself here */
for (i = 0; !found && i < clustersize/32; i++)
if (!strncmp(buffer + i * 32, filename, 12))
found = 1;
cluster = FAT[cluster];
}
while (!found && cluster != endcluster) {
readcluster(cluster); /* handle errors yourself here */
for (i = 0; !found && i < clustersize/32; i++)
if (!strncmp(buffer + i * 32, filename, 12))
found = 1;
cluster = FAT[cluster];
}
After that, you'll have the directory entry at buffer + i * 32. Of course, the indexing into the FAT is a bit of a problem. And the readcluster function is left as an exercise to the reader.
Also, please check if your memory allocation works out. From what I've seen, you are using the buffer at 7E00 both for the root directory and the FAT. I haven't read enough to know if this is a bug, but it is a code smell.
Also, why use segment 0x07c0? Isn't it easier to use segment 0? And what CPU are you targeting? The FS and GS references put it past the 8086 for sure (weren't these added with the 386 or even later?), but no 32-bit regs or variable shifts suggest otherwise
As for error handling: Why not let the user reboot, themselves? Just print "Press CTRL-ALT-DEL to reboot", then hang yourself into a loop that runs the HLT instruction with interrupts enabled. User can press CTRL-ALT-DEL, or they can turn of the computer. No int 16h required at all. As long as you have a stack set up, it'll work. Also, might want to think about continuous tone on error, in case the user of the code has no monitor and no keyboard (e.g. a server).