Re: [PATCH] seq_file: Unconditionally use vmalloc for buffer

[Kees Cook <keescook@chromium.org>]

On Mon, Mar 15, 2021 at 09:34:18AM +0100, Michal Hocko wrote:
> On Fri 12-03-21 12:55:58, Kees Cook wrote:
> > The sysfs interface to seq_file continues to be rather fragile, as seen
> > with some recent exploits[1]. Move the seq_file buffer to the vmap area
> > (while retaining the accounting flag), since it has guard pages that
> > will catch and stop linear overflows. This seems justified given that
> > seq_file already uses kvmalloc(), that allocations are normally short
> > lived, and that they are not normally performance critical.
> 
> What is the runtime effect of this change? The interface is widely used

I haven't been able to measure any differences yet, but maybe I lack
imagination about workloads that are heavy on /sys or /proc accesses.

> for many other interfaces - e.g. in proc. While from the correctness POV
> this should be OK (ish for 64b it is definitely problem for kernels with
> lowmem and limited vmalloc space). Vmalloc is also to be expected to
> regress in performance for small allocations which is the most usual
> case.

seq_file's default size is PAGE_SIZE (and just goes up by powers of 2
from there), with the rare (3 callers) exception of single_open_size(),
which for at least 1 case is always >PAGE_SIZE. (I realize PAGE_SIZE may
be considered "small" for vmalloc, but I think gaining the guard page is
worth it, given the recurring flaws we see with at least sysfs handlers.)

-Kees

>  
> > [1] https://blog.grimm-co.com/2021/03/new-old-bugs-in-linux-kernel.html
> > 
> > Signed-off-by: Kees Cook <keescook@chromium.org>
> > ---
> >  fs/seq_file.c | 10 +++++-----
> >  1 file changed, 5 insertions(+), 5 deletions(-)
> > 
> > diff --git a/fs/seq_file.c b/fs/seq_file.c
> > index cb11a34fb871..ad78577d4c2c 100644
> > --- a/fs/seq_file.c
> > +++ b/fs/seq_file.c
> > @@ -32,7 +32,7 @@ static void seq_set_overflow(struct seq_file *m)
> >  
> >  static void *seq_buf_alloc(unsigned long size)
> >  {
> > -	return kvmalloc(size, GFP_KERNEL_ACCOUNT);
> > +	return __vmalloc(size, GFP_KERNEL_ACCOUNT);
> >  }
> >  
> >  /**
> > @@ -130,7 +130,7 @@ static int traverse(struct seq_file *m, loff_t offset)
> >  
> >  Eoverflow:
> >  	m->op->stop(m, p);
> > -	kvfree(m->buf);
> > +	vfree(m->buf);
> >  	m->count = 0;
> >  	m->buf = seq_buf_alloc(m->size <<= 1);
> >  	return !m->buf ? -ENOMEM : -EAGAIN;
> > @@ -237,7 +237,7 @@ ssize_t seq_read_iter(struct kiocb *iocb, struct iov_iter *iter)
> >  			goto Fill;
> >  		// need a bigger buffer
> >  		m->op->stop(m, p);
> > -		kvfree(m->buf);
> > +		vfree(m->buf);
> >  		m->count = 0;
> >  		m->buf = seq_buf_alloc(m->size <<= 1);
> >  		if (!m->buf)
> > @@ -349,7 +349,7 @@ EXPORT_SYMBOL(seq_lseek);
> >  int seq_release(struct inode *inode, struct file *file)
> >  {
> >  	struct seq_file *m = file->private_data;
> > -	kvfree(m->buf);
> > +	vfree(m->buf);
> >  	kmem_cache_free(seq_file_cache, m);
> >  	return 0;
> >  }
> > @@ -585,7 +585,7 @@ int single_open_size(struct file *file, int (*show)(struct seq_file *, void *),
> >  		return -ENOMEM;
> >  	ret = single_open(file, show, data);
> >  	if (ret) {
> > -		kvfree(buf);
> > +		vfree(buf);
> >  		return ret;
> >  	}
> >  	((struct seq_file *)file->private_data)->buf = buf;
> > -- 
> > 2.25.1
> 
> -- 
> Michal Hocko
> SUSE Labs

-- 
Kees Cook