Matthew Wilcox <matthew@xxxxxx> writes: > On Sun, Feb 12, 2006 at 05:18:04PM +0100, Thomas Maguin wrote: >> Blame on me, a typo in the addresses. Changes should be this: >> (patch is on the way) >> >> /usr/src/linux/include/scsi/scsi.h >> +#define PLEXTOR_MODE ?? ?? ?? ?? ?? ?? ?? ?? ??0xe9 >> +#define PLEXTOR_PREC_SPEEDS ?? 0xeB >> +#define PLEXTOR_MODE2 ?? ?? ?? ?? ?? ?? ?? ??0xeD >> >> /usr/src/linux/block/scsi/scsi_ioctl.c >> +safe_for_read(PLEXTOR_MODE) >> +safe_for_read(PLEXTOR_PREC_SPEEDS) >> +safe_for_read(PLEXTOR_MODE2) > > The trouble is that these are vendor-specific commands. So they're safe > *for plextor devices*, but they may well be unsafe for other vendors. > Given the LG CD-ROM snafu [1], I'd expect people to err on the side > of caution. Sorry. The LG disaster is unrelated, they overloaded a *standard* command for their own purposes. Outright stupid, but that's LG's problem, not Linux's. This is about vendor-specified opcodes being whitelisted. I, too, consider this dangerous, unless someone desires to implement two tables in the kernel: Table 1 maps drive/model/firmware pairs to a command set table (table 2) -- this is for compression, as many drive/model/firmware patterns can share the same command whitelist Table 2 implements the actual whitelists. And if someone is going this way, perhaps the kernel should only have the mechanism and receive the tables at boot-time from a user space process (hotplug or whatever) so it doesn't need to keep a gazillion of tables in core for drives the user will never have. -- Matthias Andree - : send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
