Re: [PATCH v2] sound/oss: remove offset from load_patch callbacks

[Takashi Iwai]

At Wed, 23 Mar 2011 10:06:22 -0400,
Dan Rosenberg wrote:
> 
> On Wed, 2011-03-23 at 14:59 +0100, Takashi Iwai wrote:
> > At Wed, 23 Mar 2011 09:47:22 -0400,
> > Dan Rosenberg wrote:
> > > --- a/sound/oss/midi_synth.c
> > > +++ b/sound/oss/midi_synth.c
> > > @@ -508,16 +506,15 @@ midi_synth_load_patch(int dev, int format, const char __user *addr,
> > >  	 * been transferred already.
> > >  	 */
> > >  
> > > -	if(copy_from_user(&((char *) &sysex)[offs], &(addr)[offs], hdr_size - offs))
> > > +	if (copy_from_user(&sysex, addr, hdr_size))
> > 
> > Please correct the comment in the above as well.
> > With this change, the transfer is no longer partial.
> 
> Will do.
> 
> > 
> > > --- a/sound/oss/opl3.c
> > > +++ b/sound/oss/opl3.c
> > ...
> > > @@ -834,7 +834,7 @@ static int opl3_load_patch(int dev, int format, const char __user *addr,
> > >  	 * What the fuck is going on here?  We leave junk in the beginning
> > >  	 * of ins and then check the field pretty close to that beginning?
> > >  	 */
> > > -	if(copy_from_user(&((char *) &ins)[offs], addr + offs, sizeof(ins) - offs))
> > > +	if (copy_from_user(&ins, addr, sizeof(ins)))
> > 
> > Ditto.
> 
> Likewise.
> 
> > 
> > >  		return -EFAULT;
> > >  
> > >  	if (ins.channel < 0 || ins.channel >= SBFM_MAXINSTR)
> > > diff --git a/sound/oss/sequencer.c b/sound/oss/sequencer.c
> > > index 5ea1098..2e842cb 100644
> > > --- a/sound/oss/sequencer.c
> > > +++ b/sound/oss/sequencer.c
> > > @@ -241,7 +241,7 @@ int sequencer_write(int dev, struct file *file, const char __user *buf, int coun
> > >  				return -ENXIO;
> > >  
> > >  			fmt = (*(short *) &event_rec[0]) & 0xffff;
> > > -			err = synth_devs[dev]->load_patch(dev, fmt, buf, p + 4, c, 0);
> > > +			err = synth_devs[dev]->load_patch(dev, fmt, buf, c, 0);
> > 
> > The address must be "buf + p + 4" instead of "buf", when you omit the
> > offset argument.
> > 
> 
> Are you sure?  Previously, the copy of the header was from (buf + p + 4)
> to (dst + p + 4) with length (hdr_size - (p + 4)), and now the copy is
> from buf to dst with length hdr_size.  If I do as you suggest, then the
> copy will be from (buf + p + 4) to dst for the header.

Hm, looking back to the original code, I found that I almost forgot
the trickiness of patch-loading in OSS.  Actually, it should be like:

	err = synth_devs[dev]->load_patch(dev, fmt, buf + p, c, 0);

Here, "buf + p" points to the 4-byte event packet that is being
processed.  And, these 4 bytes are also shared with the first 4 bytes
of struct sysex_info or struct sbi_instrument.  That is,
sysex_info.key == fmt and sysex_info.device_no == dev.

The problem in the original code is that it tries to avoid extra
copying of these 4 bytes, but failed to handle correctly, by
misinterpreting offs parameter.


thanks,

Takashi
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/