[PATCH] read_file: Make it work on tftp servers which do not pass size

[Jan Weitzel]

Am Donnerstag, den 20.06.2013, 17:24 +0200 schrieb Sascha Hauer:
> On Thu, Jun 20, 2013 at 01:15:31PM +0200, Jan Weitzel wrote:
> > Am Mittwoch, den 19.06.2013, 22:58 +0200 schrieb Sascha Hauer:
> > > Some tftp servers (for example netkit-tftp) do not pass the filesize.
> > > Add a workaround for read_file which reads the file into a temporary
> > > file which then is copied to a buffer.
> > > 
> > > Signed-off-by: Sascha Hauer <s.hauer at pengutronix.de>
> > > ---
> > >  fs/fs.c      | 18 ++++++++++++++++++
> > >  fs/tftp.c    |  5 ++++-
> > >  include/fs.h |  2 ++
> > >  3 files changed, 24 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/fs/fs.c b/fs/fs.c
> > > index dc3a6e3..7046f2c 100644
> > > --- a/fs/fs.c
> > > +++ b/fs/fs.c
> > > @@ -38,10 +38,21 @@ void *read_file(const char *filename, size_t *size)
> > >  	int fd;
> > >  	struct stat s;
> > >  	void *buf = NULL;
> > > +	const char *tmpfile = "/.read_file_tmp";
> > > +	int ret;
> > >  
> > > +again:
> > >  	if (stat(filename, &s))
> > >  		return NULL;
> > >  
> > > +	if (s.st_size == FILESIZE_MAX) {
> > > +		ret = copy_file(filename, tmpfile, 0);
> > > +		if (ret)
> > > +			return NULL;
> > > +		filename = tmpfile;
> > > +		goto again;
> > > +	}
> > > +
> > >  	buf = xzalloc(s.st_size + 1);
> > >  
> > >  	fd = open(filename, O_RDONLY);
> > > @@ -56,12 +67,19 @@ void *read_file(const char *filename, size_t *size)
> > >  	if (size)
> > >  		*size = s.st_size;
> > >  
> > > +	if (filename == tmpfile)
> > > +		unlink(tmpfile);
> > > +
> > >  	return buf;
> > >  
> > >  err_out1:
> > >  	close(fd);
> > >  err_out:
> > >  	free(buf);
> > > +
> > > +	if (filename == tmpfile)
> > > +		unlink(tmpfile);
> > > +
> > >  	return NULL;
> > >  }
> > >  
> > > diff --git a/fs/tftp.c b/fs/tftp.c
> > > index 98cbb37..1c37acf 100644
> > > --- a/fs/tftp.c
> > > +++ b/fs/tftp.c
> > > @@ -597,7 +597,10 @@ static int tftp_stat(struct device_d *dev, const char *filename, struct stat *s)
> > >  		return PTR_ERR(priv);
> > >  
> > >  	s->st_mode = S_IFREG | S_IRWXU | S_IRWXG | S_IRWXO;
> > > -	s->st_size = priv->filesize;
> > > +	if (priv->filesize)
> > > +		s->st_size = priv->filesize;
> > > +	else
> > > +		s->st_size = FILESIZE_MAX;
> > Maybe we can determine the size here? Commands like "ls -l" and
> > "ubiformat" asks about the filesize via stat. I'm not sure how much
> > overheat it will be.
> 
> How do you want to do that? You would have to transfer the whole file
> first and see how big it is. That works for small files we expect to fit
> into memory like the ones read_file normally is called with. If you want
> to transfer a rootfs image it might happen that it's bigger than the
> available memory.
That's a good point. I didn't see a way for big files. But setting the
st_size to FILESIZE_MAX can cause trouble in other commands. ubiformat
only blames that is doesn't fit to eraseblock boundaries. ll -l shows a
really big size. What do you think about handle it complete in read_file
if size == 0?

Jan
> 
> Sascha
>