Re: [Qemu-devel] [PATCH 2/3] dmg: use pread

[Christoph Hellwig]

On Mon, May 10, 2010 at 12:07:40PM +0200, Kevin Wolf wrote:
> >  
> > -    info_begin=read_off(s->fd);
> > -    if(info_begin==0)
> > -   goto fail;
> > -    if(lseek(s->fd,info_begin,SEEK_SET)<0)
> > -   goto fail;
> 
> We seek to info_begin.
> 
> > -    if(read_uint32(s->fd)!=0x100)
> > -   goto fail;
> 
> Now we are at info_begin + 4
> 
> > -    if((count = read_uint32(s->fd))==0)
> > -   goto fail;
> 
> info_begin + 8
> 
> > -    info_end = info_begin+count;
> > -    if(lseek(s->fd,0xf8,SEEK_CUR)<0)
> 
> info_begin + 0x100
> 
> > +    info_begin = read_off(s->fd, offset);
> > +    if (info_begin == 0) {
> >     goto fail;
> > +    }
> > +
> > +    if (read_uint32(s->fd, info_begin) != 0x100) {
> > +        goto fail;
> > +    }
> > +
> > +    count = read_uint32(s->fd, info_begin + 4);
> > +    if (count == 0) {
> > +        goto fail;
> > +    }
> > +    info_end = info_begin + count;
> > +
> > +    offset = info_begin + 0xfc;
> 
> So, wrong offset here?

Yeah, should be 0x100.  That's what you get for quickly doing hex
calculation in your head.

> > +   if (type == 0x6d697368 && count >= 244) {
> >         int new_size, chunk_count;
> > -       if(lseek(s->fd,200,SEEK_CUR)<0)
> > -           goto fail;
> > +
> > +            offset += 4;
> 
> Isn't this needed in the else case, too?

I don't think so.  For that case we previously did a

        lseek(s->fd,count-4,SEEK_CUR)

to undo the 4 byte advance done by the read.

> > -           s->sectors[i] = last_out_offset+read_off(s->fd);
> > -           s->sectorcounts[i] = read_off(s->fd);
> > -           s->offsets[i] = last_in_offset+read_off(s->fd);
> > -           s->lengths[i] = read_off(s->fd);
> > +           read_uint32(s->fd, offset);
> 
> This read is useless. offset += 4 alone should be enough.

Thanks, fixed.

> >         /* we need to buffer, because only the chunk as whole can be
> >          * inflated. */
> >         i=0;
> >         do {
> > -           ret = read(s->fd, s->compressed_chunk+i, s->lengths[chunk]-i);
> > +           ret = pread(s->fd, s->compressed_chunk+i, s->lengths[chunk]-i,
> > +                            s->offsets[chunk]);
> 
> This is in a loop, whereas the lseek was outside the loop. From the
> second iteration on you'll repeat the first read instead of advancing.

You're right.  The EINTR check confused me an I took this for just
retrying reads on EINTR.  Now this code i quite nasty for error returns
except EINTR because we'll subtract one from the i loop iteration,
yikes.  I'll just reuse the i variable to keep the same kind of bug
for both sides of the equation.

God, do I hate this code..