Why is there read SCSI command during a sys_write()

Question

I want to understand how a sys_write on a scsi device works. Therefore I wrote a simple test program, and added in the issue function of the device driver some debug printk() to print the ata_queued_cmd->scsicmd->cmnd.

int main() {
    int w;
    char buffer[512] = "test string with 512 byte";

    // OPEN
    int fd = open("/dev/sdd", O_WRONLY | O_SYNC);

    // WRITE
    w = write(fd,buffer,sizeof(buffer));

    return 0;
}

In /var/log/messages I can see the following issued commands:

  scsicmd->cmnd
------------------------------
 0x28 | READ(10)
 0x2A | WRITE(10)
 0x35 | SYNCHRONIZE CACHE (10)

The 512 Byte I want to write should be the blocksize. This is how I get this value:

root$ blockdev --getsize /dev/sdd # returns size in sectors
488397168
root$ blockdev --getsize64 /dev/sdd # returns size in sectors
250059350016
root$ python -c "print 250059350016.0/488397168"
512.0

It's very important for me, that the data is power-failure safe written. That's the reason why I want to wait for the reuturn with O_SYNC. Now I have some questions:

• Why is there a READ before the WRITE?
• Does the SYNCHRONIZE CACHE guarantee me, if the data is in the internal cache of the drive? Or is there more I could do to have a secure write?

Thank you very much :)

Answer 1

OK I figured it out now. They read it to load the data into an internal Kernel buffer, so when you have multiple writes/reads, the data is already buffered.