On Oct 12, 2020, at 9:57 AM, Arthur Krewat <krewat@kilonet.net> wrote:
>
> On 10/11/2020 9:56 PM, Warner Losh wrote:
>> unaligned I/O on the block device
> Sorry, I have to laugh... isn't that an oxymoron? ;)
Actually there are no *block* devices in FreeBSD.
And while raw device files such as /dev/r<blkdev>
no longer exist, it is the block devices (with
their buffering etc.) that are gone. Only raw devices
exist now.
There's still the buffer cache that does do buffering on the devices... but it's a different kind of buffering than traditional kernels. but you're right, only the cdevs are plumbed down, not the bdevs.
And FreeBSD raw disk device drivers don't allow
unaligned I/O access. Nor should they paper over
what is not allowed by the underlying device. Try
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
unsigned char buf[1024];
int main(int c, char**v) {
char* a = c > 1? v[1] : "1";
off_t off = (off_t)strtoull(a, 0, 0);
ssize_t r = pread(0, buf, sizeof buf, off);
if (r < 0) { perror("read"); return 1; }
for (int i = 0; i < 4; i++) printf(" %02x", buf[i]);
printf("\n");
return 0;
}
I meant something different about 'unaligned' access. I mean that the buffers you read the data into needn't be page aligned. That causes a fair amount of hair in the lower layers where I spend much of my time...
You are correct you have to read them on an LBA boundary.
Warner