Marc SCHAEFER wrote:
>
> The question of the differences between /dev/sg write()/read() interface
> and SCSI_IOCTL_SEND_COMMAND was raised in the amanda-users mailing-list.
>
> I answered with the following. Any comments ? I will do a summary
> to the amanda-users mailing-list.
The big difference is that SCSI_IOCTL_SEND_COMMAND has a PAGE_SIZE limit (2048
bytes on i386), which means it cannot be used for large tape changers and
large optical juke boxes because the results of a READ_ELEMENT_STATUS SCSI
command would exceed 2048 bytes. That's why I moved away from the
SCSI_IOCTL_SEND_COMMAND to the /dev/sg interface for the version of 'mtx' that
I did for large tape changers and libraries (see:
ftp://ftp.estinc.com/pub/unsupported/mtxl.README.html ).
The upside is that SCSI_IOCTL_SEND_COMMAND appears to be more reliable than
the /dev/sg interface at handling the case of "shit happens". If my program
segfaults between the write and the read with the /dev/sg interface, for
example, half the time I appear to have hosed my entire SCSI subsystem and end
up rebooting because the Adaptec driver is stuck in a reset loop ("trying....
trying again... failed."). In production use (as vs me hacking on it and
breaking it :-) this should never happen, but as the saying goes, the best
laid plans ...
> The LINUX_CHG one uses the generic interface (/dev/sg), but it seems
> it only works in 2.0.x (sg was changed, theoretically in a
> backwards-compatible way in 2.2.x ...)
http://www.torque.net/sg/
Especially see the 'sgtools' package at that site, which has a large amount of
sample code using both SCSI_IOCTL_SEND_COMMAND and equivalent code that uses
the /dev/sg interface.
--
Eric Lee Green [EMAIL PROTECTED]
Software Engineer Visit our Web page:
Enhanced Software Technologies, Inc. http://www.estinc.com/
(602) 470-1115 voice (602) 470-1116 fax
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
